-
Notifications
You must be signed in to change notification settings - Fork 0
Expand file tree
/
Copy pathpapers.bib
More file actions
1473 lines (1386 loc) · 196 KB
/
papers.bib
File metadata and controls
1473 lines (1386 loc) · 196 KB
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
---
---
@string{aps = {American Physical Society,}}
@article{Beljaards:2026,
abbr = {},
bibtex_show = {true},
author = {Beljaards, Laurens and Nagtegaal, Martijn and Rao, Chinmay and Dong, Yiming and van Osch, Matthias J.P. and Pezzotti, Nicola and Doneva, Mariya and Staring, Marius},
title = {DEEP-DISORDER: Motion Correction in 3D MRI via Segment Reconstruction and Registration},
journal = {NMR in Biomedicine},
volume = {},
number = {},
pages = {},
year = {2026},
pdf = {2026_j_NMRbiomed.pdf},
html = {},
arxiv = {},
code = {},
abstract = {3D MR image acquisition is inherently time-intensive, rendering it susceptible to patient motion during scanning. This may introduce significant blurring and artifacts, potentially necessitating re-acquisition.<br>We propose a modular framework to retrospectively correct for intra-scan motion in 3D brain MRI, without active motion tracking. Serving as the backbone of our approach is an existing distributed and incoherent sampling scheme (DISORDER), combined with a fast network trained for highly undersampled reconstruction. This enables approximate reconstructions of anatomy after every few seconds, using only a tiny fraction of k-space data (<2%). While these reconstructions are only approximate, we postulate they are sufficient to estimate motion patterns at said temporal resolution. Groupwise registration, notable for its elimination of registration bias, is utilized for estimating rigid motion parameters, which are leveraged to reconstruct the measured data with reduced motion artifacts.<br>The approach was evaluated on 94 retrospectively and 3 prospectively motion-corrupted in vivo 3D T1-weighted brain MRI acquisitions. The estimated motion parameters matched the known retrospective motion with 0.06 mm and 0.13° accuracy, resulting in an improvement in reconstruction quality from 0.942±0.026 to 0.992±0.003 SSIM for the retrospective scans. The prospective scans improved from 0.915±0.024 to 0.936±0.014 SSIM after correction in the case of gradual motion and from 0.764±0.008 to 0.923±0.011 SSIM for extreme motion.<br>In conclusion, the proposed approach, that is free of external tracking devices or navigators, successfully estimated and corrected 3D motion between small sub-portions of a scan. This resulted in vastly improved image quality, making volumetric MRI substantially more tolerant to motion.},
}
@article{Lyu:2026,
abbr = {},
bibtex_show = {true},
author = {Lyu, Donghang and Staring, Marius and van Osch, Matthias J. P. and Doneva, Mariya and Lamb, Hildo J. and Pezzotti, Nicola},
title = {Convolutional recurrent U-net for cardiac cine MRI reconstruction via effective spatio-temporal feature exploitation},
journal = {Medical Physics},
volume = {53},
number = {1},
pages = {e70245},
year = {2026},
pdf = {2026_j_MP.pdf},
html = {https://dx.doi.org/10.1002/mp.70245},
arxiv = {},
code = {https://github.com/dong845/CRUNet-MR/tree/main},
abstract = {<b>Background:</b> Cardiac Cine Magnetic Resonance Imaging (MRI) provides dynamic visualization of the heart's structure and function but is hindered by slow acquisition, requiring repeated breath-holds that challenge sick patients. Accelerated imaging can mitigate these issues but potentially reduce spatial and temporal resolution. Therefore, innovative approaches are essential to ensure effective performance under high acceleration conditions. Deep learning-based reconstruction methods show promise in enhancing image quality from highly undersampled data, accelerating scans while maintaining diagnostic accuracy. However, they often fail to effectively exploit the spatio-temporal features inherent to cine MRI, which are essential for accurate reconstruction, thereby leaving room for further improvement.<br><b>Purpose:</b> We aim to more effectively exploit the spatio-temporal features inherent in cine MRI sequences by integrating convolutional recurrent operations with a U-Net architecture, enhancing the reconstruction performance of cine MRI.<br><b>Methods:</b> We developed a new deep learning model called CRUNet-MR that enhances the extraction of spatio-temporal features by combining convolutional recurrent operations with a U-Net structure. This design ensures continuous extraction of temporal features while fusing fine-grained spatial details with high-level semantic information. Furthermore, dilated convolutions are incorporated to expand the spatial receptive field, and appropriate combinations of dilation factors are explored to further enhance overall performance.<br><b>Results:</b> Training, validation, and testing were performed on the public CMRxRecon2023 dataset, using two views and four acceleration factors ranging from 4 to 24 with the given Auto-Calibration Signal (ACS) area. The dataset consists of 120 subjects for training, 60 for validation, and 120 for testing. In general, the proposed CRUNet-MR shows statistically significant differences with benchmark models and consistently outperforms them, particularly showcasing better reconstruction quality in dynamic regions, highlighting its effective extraction of spatio-temporal features. Ablation studies further validated the design choices of CRUNet-MR. The model demonstrated strong reconstruction performance, achieving an average SSIM of 0.986 at an acceleration factor of 4 and 0.971 at a factor of 8 across both views. Furthermore, CRUNet-MR was validated on a small in-house LUMC dataset, showing its generalization capability and rapid adaptability through fine-tuning.<br><b>Conclusions:</b> The proposed CRUNet-MR model is well-suited for cine MRI reconstruction, effectively leveraging spatio-temporal features to reconstruct high-quality images, especially in dynamic cardiac regions. This capability highlights its potential to support higher acceleration factors, enabling faster and more patient-friendly cardiac imaging.},
}
@article{Chen:2025,
abbr = {},
bibtex_show = {true},
author = {Chen, Yunjie and Weber, Rianne and Neve, Olaf M. and Romeijn, Stephan R. and Hensen, Erik F. and Wolterink, Jelmer M. and Tao, Qian and Staring, Marius and Verbist, Berit M.},
title = {A deep learning model to reduce agent dose for contrast-enhanced MRI of the cerebellopontine angle cistern},
journal = {European Radiology},
volume = {},
pages = {},
year = {2025},
pdf = {2025_j_ER.pdf},
html = {https://doi.org/10.1007/s00330-025-12187-8},
arxiv = {2511.20926},
code = {},
abstract = {<b>Objectives:</b> To evaluate a deep learning (DL) model for reducing the agent dose of contrast-enhanced T1-weighted MRI (T1ce) of the cerebellopontine angle (CPA) cistern.<br><b>Materials and methods:</b> In this multi-center retrospective study, T1 and T1ce of vestibular schwannoma (VS) patients were used to simulate low-dose T1ce with varying reductions of contrast agent dose. DL models were trained to restore standard-dose T1ce from the low-dose simulation. The image quality and segmentation performance of the DL-restored T1ce were evaluated. A head and neck radiologist was asked to rate DL-restored images in multiple aspects, including image quality and diagnostic characterization.<br><b>Results:</b> 203 MRI studies from 72 VS patients (mean age, 58.51 ± 14.73, 39 men) were evaluated. As the input dose increased, the structural similarity index measure of the restored T1ce increased from 0.639 ± 0.113 to 0.993 ± 0.009, and the peak signal-to-noise ratio increased from 21.6 ± 3.73 dB to 41.4 ± 4.84 dB. At 10% input dose, using DL-restored T1ce for segmentation improved the Dice from 0.673 to 0.734, the 95% Hausdorff distance from 2.38 mm to 2.07 mm, and the average surface distance from 1.00 mm to 0.59 mm. Both DL-restored T1ce from 10% and 30% input doses showed excellent image quality (3.09 ± 0.811 and 3.23 ± 0.685), with the latter being considered more informative (3.81 ± 0.664).<br><b>Conclusion:</b> The DL model improved the image quality of low-dose MRI of the CPA cistern, which makes lesion detection and diagnostic characterization possible with 10% - 30% of the standard dose.<br><b>Key points</b><br><b>Question</b> Deep learning models that aid in the reduction of contrast agent dose are not extensively evaluated for MRI of the cerebellopontine angle cistern.<br><b>Finding</b> Deep learning models restored the low-dose MRI of the cerebellopontine angle cistern, yielding images sufficient for vestibular schwannoma diagnosis and management.<br><b>Clinical relevance statement</b> Deep learning models make it possible to reduce the use of gadolinium-based contrast agents for contrast-enhanced MRI of the cerebellopontine angle cistern.},
}
@article{VanDerValk:2025,
abbr = {},
bibtex_show = {true},
author = {van der Valk, Viktor and Atsma, Douwe and Scherptong, Roderick and Staring, Marius},
title = {Explainable ECG analysis by explicit information disentanglement with VAEs},
journal = {IEEE Transactions on Biomedical Engineering},
volume = {},
number = {},
pages = {},
year = {2025},
pdf = {2025_j_TBME.pdf},
html = {https://doi.org/10.1109/TBME.2025.3631143},
arxiv = {},
code = {},
abstract = {<b>Objective:</b> The interpretation of electrocardiogram (ECG) signals is vital for diagnosis of cardiac conditions. Traditional methods rely on expert knowledge, which is time consuming, costly and potentially misses subtle features. AI has shown promise in ECG interpretation, but clinically desired model explainability is often lacking in literature.<br><b>Methods:</b> We introduce an explainable AI method for ECG classification by partitioning the variational autoencoder (VAE) latent space into a label-specific and a non-label-specific subset. By optimizing both subsets for signal reconstruction and one subset also for prediction while constraining the other from learning label-specific information with an adversarial network, the latent space is disentangled in a supervised manner. This latent space is leveraged to create enhanced visualizations for ECG feature interpretation by means of attribute manipulation. As a proof of concept, we predict the left ventricular function (LVF), a critical prognostic determinant in cardiac disease, from the ECG.<br><b>Results:</b> Our study demonstrates the effective segregation of LVF-specific information within a single dimension of the VAE latent space, without compromising classification performance. We show that the proposed model improves state-of-the-art VAE methods (AUC 0.832 vs. 0.790, F1 0.688 vs. 0.640) in prediction and performs comparable to ground truth LVF (concordance 0.72 vs.0.72) in predicting survival.<br><b>Conclusion:</b> The model facilitates the interpretation of LVF predictions by providing visual context to ECG signals, offering a general explainable and predictive AI method.<br><b>Significance:</b> Our explainable AI model can potentially reduce time and expertise required for ECG analysis.},
}
@article{Li:2025,
abbr = {TMI},
bibtex_show = {true},
author = {Li, Tianran and Staring, Marius and Qiao, Yuchuan},
title = {Efficient Large-Deformation Medical Image Registration via Recurrent Dynamic Correlation},
journal = {IEEE Transactions on Medical Imaging},
volume = {},
number = {},
pages = {},
year = {2025},
pdf = {2025_j_TMI.pdf},
html = {https://doi.org/10.1109/TMI.2025.3630584},
arxiv = {2510.22380},
code = {},
abstract = {Deformable image registration estimates voxel-wise correspondences between images through spatial transformations, and plays a key role in medical imaging. While deep learning methods have significantly reduced runtime, efficiently handling large deformations remains a challenging task. Convolutional networks aggregate local features but lack direct modeling of voxel correspondences, promoting recent works to explore explicit feature matching. Among them, voxel-to-region matching is more efficient for direct correspondence modeling by computing local correlation features within neighbourhoods, while region-to-region matching incurs higher redundancy due to excessive correlation pairs across large regions. However, the inherent locality of voxel-to-region matching hinders the capture of long-range correspondences required for large deformations. To address this, we propose a Recurrent Correlation-based framework that dynamically relocates the matching region toward more promising positions. At each step, local matching is performed with low cost, and the estimated offset guides the next search region, supporting efficient convergence toward large deformations. In addition, we uses a lightweight recurrent update module with memory capacity and decouples motion-related and texture features to suppress semantic redundancy. We conduct extensive experiments on brain MRI and abdominal CT datasets under two settings: with and without affine pre-registration. Results show our method exhibits a strong accuracy-computation trade-off, surpassing or matching the state-of-the-art performance. For example, it achieves comparable performance on the non-affine OASIS dataset, while using only 9.5% of the FLOPs and running 96% faster than RDP, a representative high-performing method.},
}
@article{Chaves-de-Plaza:2026,
abbr = {TVCG},
bibtex_show = {true},
author = {Chaves-de-Plaza, Nicolas and Raidou, Renata G. and Mody, Prerak P. and Staring, Marius and van Egmond, Ren{\'e}; and Vilanova, Anna and Hildebrandt, Klaus},
title = {LoGCC: Local-to-Global Correlation Clustering for Scalar Field Ensembles},
journal = {IEEE Transactions on Visualization and Computer Graphics},
volume = {32},
number = {2},
pages = {2260 -- 2271},
year = {2026},
month = {February},
pdf = {2026_j_TVCG.pdf},
html = {https://doi.org/10.1109/TVCG.2025.3630550},
arxiv = {},
code = {},
abstract = {Correlation clustering (CC) offers an effective approach to analyze scalar field ensembles by detecting correlated regions and consistent structures, enabling the extraction of meaningful patterns. However, existing CC methods are computationally expensive, making them impractical for both interactive analysis and large-scale scalar fields. We introduce the Local-to-Global Correlation Clustering (LoGCC) framework, which accelerates pivot-based CC by leveraging the spatial structure of scalar fields and the weak transitivity of correlation. LoGCC operates in two stages: a local step that uses the neighborhood graph of the scalar field's spatial domain to build highly correlated local clusters, and a global step that merges them into global clusters. We implement the LoGCC framework for two well-known pivot-based CC methods, Pivot and CN-Pivot, demonstrating its generality. Our evaluation using synthetic and real-world meteorological and medical image segmentation datasets shows that LoGCC achieves speedups-up to 15x for Pivot and 200x for CN-Pivot-and improved scalability to larger scalar fields, while maintaining cluster quality. These contributions broaden the applicability of correlation clustering in large-scale and interactive analysis settings.},
}
@article{Du2025,
abbr = {},
bibtex_show = {true},
title = {Predictive value of aorta enhancement on computed tomographic pulmonary angiography in pulmonary embolism},
author = {Du, Quiyu and ter Haar, S.N.M. and Jia, Jingnan and Kroft, Lucia J.M. and Staring, Marius and Klok, F.A. and Stoel, Berend C.},
journal = {PLOS ONE},
volume = {20},
number = {10},
pages = {1 -- 15},
month = {October},
year = {2025},
pdf = {2025_j_PONE.pdf},
html = {https://doi.org/10.1371/journal.pone.0335055},
arxiv = {},
code = {},
abstract = {<b>Background:</b> Pulmonary embolism (PE) is a life-threatening condition requiring prompt diagnosis and treatment. Visual assessment of computed tomographic pulmonary angiography (CTPA) is the first-choice diagnostic tool. New imaging biomarkers could provide additional prognostic information for improved risk stratification. We hypothesized in this exploratory study, that contrast enhancement patterns in the aorta may contain such information.<br><b>Methods:</b> CTPA scans of 93 acute PE patients were analyzed retrospectively. Firstly, the aorta was segmented automatically by TotalSegmentator and its centerline was extracted. Subsequently, lines were fitted on intensities within a region of interest perpendicularly to the aorta centerline, from which three parameters were extracted: mean intensity, proximal intensity and contrast gradient. After confounder analysis, logistic regression with forward selection evaluated the predictive value of these parameters for 12 adverse outcomes (six short-term and six long-term).<br><b>Results:</b> Lung volume, aorta dimension and contrast delay were considered as possible confounders but were not selected by forward selection. Logistic regression (n = 93) showed that a less steep contrast gradient (decreasing by 10 Hounsfield unit/%) was associated with a reduction in odds of the following short-term adverse outcomes: 48.1% for intensive care unit admission (odds ratio [OR] = 0.519, 95% confidence interval [CI]: 0.306-0.804), 29.3% for oxygen therapy >24 hours (OR = 0.707, 95% CI: 0.496-0.976), 60.6% for reperfusion therapy (OR = 0.394, 95% CI: 0.178-0.682), 57.5% for vasopressor therapy (OR = 0.425, 95% CI: 0.194-0.741), and 50.2% for PE-related death (OR = 0.498, 95% CI: 0.246-0.915). No significant associations were found with long-term outcomes.<br><b>Conclusions:</b> The aorta contrast gradient, automatically quantified from CTPA, is a relevant adjunctive predictor for short-term outcomes in PE patients. Long-term outcomes, however, could not be predicted by aorta measurement. This pilot study provides initial insights into predictive value of aorta enhancement, stimulating further exploration with external data.},
}
@article{VanDerLoo2025,
abbr = {},
bibtex_show = {true},
title = {Large Language Models for Structured Cardiovascular Data Extraction: A Foundation for Scalable Research and Clinical Applications},
author = {van der Loo, Wouter and van der Valk, Viktor and van den Broek, Tim and Atsma, Douwe and Staring, Marius and Scherptong, Roderick},
journal = {European Heart Journal - Digital Health},
volume = {7},
number = {2},
pages = {},
month = {March},
year = {2025},
pdf = {2025_j_EHJ-DH.pdf},
html = {https://doi.org/10.1093/ehjdh/ztaf127},
arxiv = {},
code = {},
abstract = {<b>Background:</b> Automated extraction of information from cardiac reports would benefit both clinical reporting and research. Large language models (LLMs) hold promise for such automation, but their clinical performance and practical implementation across various computational environments remain unclear.<br><b>Objectives:</b> To evaluate the feasibility and performance of LLM-based classification of echocardiogram and invasive coronary angiography (ICA) reports, using real-world clinical data across local, high-performance computing and cloud-based platforms.<br><b>Methods:</b> The angiography and echocardiography reports of 1000 patients, admitted with acute coronary syndrome, were labeled for multiple key diagnostic elements, including left ventricular function (LVF), culprit vessel and acute occlusions. Report classification models were developed using LLMs via i. prompt-based and ii. fine-tuning approaches. Performance was assessed across different model types and compute infrastructures, with attention to class imbalance, ambiguous label annotations and implementation costs.<br><b>Results:</b> LLMs demonstrated strong performance in extracting structured diagnostic information from cardiac reports. Cloud-based models (such as GPT-4o) achieved the highest accuracy (0.87 for culprit vessel and 1.0 for LVF) and generalizability, but also smaller models run on a local high performance cluster (HPC) achieved reasonable accuracy, especially for less complex tasks (0.634 for culprit vessel and 0.984 for LVF). Classification was feasible with minimal preprocessing, enabling potential integration into electronic health record systems or research pipelines. Class imbalance, reflective of real-world prevalence, had a greater impact on fine-tuning approaches.<br><b>Conclusions:</b> LLMs can reliably classify structured cardiology reports across diverse compute infrastructures. Their accuracy and adaptability support their use in clinical and research settings, particularly for scalable report structuring and dataset generation.},
}
@article{Elmahdy2025,
abbr = {},
bibtex_show = {true},
title = {CMRINet: Joint Groupwise Registration and Segmentation for Cardiac Function Quantification from Cine-MRI},
author = {Elmahdy, Mohamed S. and Staring, Marius and de Koning, Patrick J. H. and Alabed, Samer and Salehi, Mahan and Alandejani, Faisal and Sharkey, Michael and Aldabbagh, Ziad and Swift, Andrew J. and van der Geest, Rob J.},
journal = {arXiv},
volume = {},
pages = {},
month = {},
year = {2025},
pdf = {2025_j_arxiv.pdf},
html = {https://arxiv.org/abs/2505.16452},
arxiv = {2505.16452},
code = {},
abstract = {Accurate and efficient quantification of cardiac function is essential for the estimation of prognosis of cardiovascular diseases (CVDs). One of the most commonly used metrics for evaluating cardiac pumping performance is left ventricular ejection fraction (LVEF). However, LVEF can be affected by factors such as inter-observer variability and varying pre-load and after-load conditions, which can reduce its reproducibility. Additionally, cardiac dysfunction may not always manifest as alterations in LVEF, such as in heart failure and cardiotoxicity diseases. An alternative measure that can provide a relatively load-independent quantitative assessment of myocardial contractility is myocardial strain and strain rate. By using LVEF in combination with myocardial strain, it is possible to obtain a thorough description of cardiac function. Automated estimation of LVEF and other volumetric measures from cine-MRI sequences can be achieved through segmentation models, while strain calculation requires the estimation of tissue displacement between sequential frames, which can be accomplished using registration models. These tasks are often performed separately, potentially limiting the assessment of cardiac function. To address this issue, in this study we propose an end-to-end deep learning (DL) model that jointly estimates groupwise (GW) registration and segmentation for cardiac cine-MRI images. The proposed anatomically-guided Deep GW network was trained and validated on a large dataset of 4-chamber view cine-MRI image series of 374 subjects. A quantitative comparison with conventional GW registration using elastix and two DL-based methods showed that the proposed model improved performance and substantially reduced computation time.},
}
@article{Lyu2025,
abbr = {},
bibtex_show = {true},
title = {MCP-MedSAM: A Powerful Lightweight Medical Segment Anything Model Trained with a Single GPU in Just One Day},
author = {Lyu, Donghang and Gao, Ruochen and Staring, Marius},
journal = {The Journal of Machine Learning for Biomedical Imaging},
volume = {3},
pages = {135 -- 151},
month = {May},
year = {2025},
pdf = {2025_j_MELBA.pdf},
html = {https://doi.org/10.59275/j.melba.2025-4849},
arxiv = {},
code = {https://github.com/dong845/MCP-MedSAM},
abstract = {Medical image segmentation involves partitioning medical images into meaningful regions, with a focus on identifying anatomical structures and lesions. It has broad applications in healthcare, and deep learning methods have enabled significant advancements in automating this process. Recently, the introduction of the Segmentation Anything Model (SAM), the first foundation model for segmentation task, has prompted researchers to adapt it for the medical domain to improve performance across various tasks. However, SAM's large model size and high GPU requirements hinder its scalability and development in the medical domain. To address these challenges, research has increasingly focused on lightweight adaptations of SAM to reduce its parameter count, enabling training with limited GPU resources while maintaining competitive segmentation performance. In this work, we propose MCP-MedSAM, a powerful and lightweight medical SAM model designed to be trainable on a single A100 GPU with 40GB of memory within one day while delivering superior segmentation performance. Recognizing the significant internal differences between modalities and the need for direct segmentation target information within bounding boxes, we introduce two kinds of prompts: the modality prompt and the content prompt. After passing through the prompt encoder, their embedding representations can further improve the segmentation performance by incorporating more relevant information without adding significant training overhead. Additionally, we adopt an effective modality-based data sampling strategy to address data imbalance between modalities, ensuring more balanced performance across all modalities. Our method was trained and evaluated using a large-scale challenge dataset, compared to top-ranking methods on the challenge leaderboard, MCP-MedSAM achieved superior performance while requiring only one day of training on a single GPU. The code is publicly available at <a href="https://github.com/dong845/MCP-MedSAM">https://github.com/dong845/MCP-MedSAM</a>.},
}
@article{Gao2025,
abbr = {},
bibtex_show = {true},
title = {On Factors that Influence Deep Learning-Based Dose Prediction of Head and Neck Tumors},
author = {Gao, Ruochen and Mody, Prerak and Rao, Chinmay S. and Dankers, Frank and Staring, Marius},
journal = {Physics in Medicine and Biology},
volume = {70},
number = {11},
pages = {115006},
month = {},
year = {2025},
pdf = {2025_j_PMB.pdf},
html = {https://doi.org/10.1088/1361-6560/adcfeb},
arxiv = {},
code = {https://github.com/RuochenGao/HaN-DosePrediction},
abstract = {<i>Objective.</i> This study investigates key factors influencing deep learning-based dose prediction models for head and neck cancer radiation therapy (RT). The goal is to evaluate model accuracy, robustness, and computational efficiency, and to identify key components necessary for optimal performance.<br><i>Approach.</i> We systematically analyze the impact of input and dose grid resolution, input type, loss function, model architecture, and noise on model performance. Two datasets are used: a public dataset (OpenKBP) and an in-house clinical dataset (LUMC). Model performance is primarily evaluated using two metrics: dose score and dose-volume histogram (DVH) score.<br><i>Main results.</i> High-resolution inputs improve prediction accuracy (dose score and DVH score) by 8.6-13.5\% compared to low resolution. Using a combination of CT, planning target volumes (PTVs), and organs-at-risk (OARs) as input significantly enhances accuracy, with improvements of 57.4-86.8\% over using CT alone. Integrating mean absolute error (MAE) loss with value-based and criteria-based DVH loss functions further boosts DVH score by 7.2-7.5\% compared to MAE loss alone. In the robustness analysis, most models show minimal degradation under Poisson noise (0-0.3 Gy) but are more susceptible to adversarial noise (0.2-7.8 Gy). Notably, certain models, such as SwinUNETR, demonstrate superior robustness against adversarial perturbations.<br><i>Significance.</i> These findings highlight the importance of optimizing deep learning models and provide valuable guidance for achieving more accurate and reliable radiotherapy dose prediction.},
}
@article{Verheijen2025,
abbr = {},
bibtex_show = {true},
title = {Artificial Intelligence for Segmentation and Classification in Lumbar Spinal Stenosis: an overview of current methods},
author = {Verheijen, E.J.A. and Kapogiannis, T. and Munteh, D. and Chabros, J. and Staring, M. and Smith, T.R. and Vleggeert-Lankamp, C.L.A.},
journal = {European Spine Journal},
volume = {34},
pages = {1146 -- 1155},
month = {March},
year = {2025},
pdf = {2025_j_ESJ.pdf},
html = {https://doi.org/10.1007/s00586-025-08672-9},
arxiv = {},
code = {},
abstract = {<b>Purpose:</b> Lumbar spinal stenosis (LSS) is a frequently occurring condition defined by narrowing of the spinal or nerve root canal due to degenerative changes. Physicians use MRI scans to determine the severity of stenosis, occasionally complementing it with X-ray or CT scans during the diagnostic work-up. However, manual grading of stenosis is time-consuming and induces inter-reader variability as a standardized grading system is lacking. Machine Learning (ML) has the potential to aid physicians in this process by automating segmentation and classification of LSS. However, it is unclear what models currently exist to perform these tasks.<br><b>Methods:</b> A systematic review of literature was performed by searching the Cochrane Library, Embase, Emcare, PubMed, and Web of Science databases for studies describing an ML-based algorithm to perform segmentation or classification of the lumbar spine for LSS. Risk of bias was assessed through an adjusted version of the Newcastle-Ottawa Quality Assessment Scale that was more applicable to ML studies. Qualitative analyses were performed based on type of algorithm (conventional ML or Deep Learning (DL)) and task (segmentation or classification).<br><b>Results:</b> A total of 27 articles were included of which nine on segmentation, 16 on classification and 2 on both tasks. The majority of studies focused on algorithms for MRI analysis. There was wide variety among the outcome measures used to express model performance. Overall, ML algorithms are able to perform segmentation and classification tasks excellently. DL methods tend to demonstrate better performance than conventional ML models. For segmentation the best performing DL models were U-Net based. For classification U-Net and unspecified CNNs powered the models that performed the best for the majority of outcome metrics. The number of models with external validation was limited.<br><b>Conclusion:</b> DL models achieve excellent performance for segmentation and classification tasks for LSS, outperforming conventional ML algorithms. However, comparisons between studies are challenging due to the variety in outcome measures and test datasets. Future studies should focus on the segmentation task using DL models and utilize a standardized set of outcome measures and publicly available test dataset to express model performance. In addition, these models need to be externally validated to assess generalizability.},
}
@article{Rezaei2025,
abbr = {},
bibtex_show = {true},
title = {Bridging Gaps with Computer Vision: AI in (Bio)Medical Imaging and Astronomy},
author = {Rezaei, Samira and Chegeni, Amirmohammad and Javadpour, Amir and VafaeiSadr, Alireza and Cao, Lu and Rottgering, Huub and Staring, Marius},
journal = {Astronomy and Computing},
volume = {51},
pages = {100921},
month = {April},
year = {2025},
pdf = {2025_j_AC.pdf},
html = {https://doi.org/10.1016/j.ascom.2024.100921},
arxiv = {},
code = {},
abstract = {This paper explores how artificial intelligence (AI) techniques can address common challenges in astronomy and (bio)medical imaging. It focuses on applying convolutional neural networks (CNNs) and other AI methods to tasks such as image reconstruction, object detection, anomaly detection, and generative modeling. Drawing parallels between domains like MRI and radio astronomy, the paper highlights the critical role of AI in producing high-quality image reconstructions and reducing artifacts. Generative models are examined as versatile tools for tackling challenges such as data scarcity and privacy concerns in medicine, as well as managing the vast and complex datasets found in astrophysics. Anomaly detection is also discussed, with an emphasis on unsupervised learning approaches that address the difficulties of working with large, unlabeled datasets. Furthermore, the paper explores the use of reinforcement learning to enhance CNN performance through automated hyperparameter optimization and adaptive decision-making in dynamic environments. The focus of this paper remains strictly on AI applications, without addressing the synergies between measurement techniques or the core algorithms specific to each field.},
}
@article{ChavezDePlaza2025,
abbr = {},
bibtex_show = {true},
title = {Implementation of Delineation Error Detection Systems in Time-Critical Radiotherapy: Do AI-Supported Optimization and Human Preferences Meet?},
author = {Chaves-de-Plaza, Nicolas F. and Mody, Prerak and Hildebrandt, Klaus and Staring, Marius and Astreinidou, Eleftheria and de Ridder, Mischa and de Ridder, Huib and Vilanova, Anna and van Egmond, Rene},
journal = {Cognition, Technology & Work},
volume = {27},
pages = {41 -- 57},
month = {June},
year = {2025},
pdf = {2025_j_CTW.pdf},
html = {https://doi.org/10.1007/s10111-024-00784-4},
arxiv = {},
code = {},
abstract = {Artificial Intelligence (AI)-based auto-delineation technologies rapidly delineate multiple structures of interest like organs-at-risk and tumors in 3D medical images, reducing personnel load and facilitating time-critical therapies. Despite its accuracy, the AI may produce flawed delineations, requiring clinician attention. Quality assessment (QA) of these delineations is laborious and demanding. Delineation error detection systems (DEDS) aim to aid QA, yet questions linger about potential challenges to their adoption and time-saving potential. To address these queries, we first conducted a user study with two clinicians from Holland Proton Therapy Center, a Dutch cancer treatment center. Based on the study's findings about the clinicians' error detection workflows with and without DEDS assistance, we developed a simulation model of the QA process, which we used to assess different error detection workflows on a retrospective cohort of 42 head and neck cancer patients. Results suggest possible time savings, provided the per-slice analysis time stays close to the current baseline and trading-off delineation quality is acceptable. Our findings encourage the development of user-centric delineation error detection systems and provide a new way to model and evaluate these systems' potential clinical value.},
}
@article{Malimban2024,
abbr = {},
bibtex_show = {true},
title = {A simulation framework for preclinical proton irradiation workflow},
author = {Malimban, Justin and Ludwig, Felix and Lathouwers, Danny and Staring, Marius and Verhaegen, Frank and Brandenburg, Sytze},
journal = {Physics in Medicine and Biology},
volume = {69},
pages = {215040},
month = {},
year = {2024},
pdf = {2024_j_PMB.pdf},
html = {https://doi.org/10.1088/1361-6560/ad897f},
arxiv = {},
code = {},
abstract = {<b>Objective:</b> The integration of proton beamlines with X-ray imaging/irradiation platforms has opened up possibilities for image-guided Bragg peak irradiations in small animals. Such irradiations allow selective targeting of normal tissue substructures and tumours. However, their small size and location pose challenges in designing experiments. This work presents a simulation framework useful for optimizing beamlines, imaging protocols, and design of animal experiments. The usage of the framework is demonstrated, mainly focusing on the imaging part.<br><b>Approach:</b> The fastCAT toolkit was modified with Monte Carlo (MC)-calculated primary and scatter data of a small animal imager for the simulation of micro-CT scans. The simulated CT of a mini-calibration phantom from fastCAT was validated against a full MC TOPAS CT simulation. A realistic beam model of a preclinical proton facility was obtained from beam transport simulations to create irradiation plans in matRad. Simulated CT images of a digital mouse phantom were generated using single-energy CT (SECT) and dual-energy CT (DECT) protocols and their accuracy in proton stopping power ratio (SPR) estimation and their impact on calculated proton dose distributions in a mouse were evaluated.<br><b>Main Results:</b> The CT numbers from fastCAT agree within 11 HU with TOPAS except for materials at the centre of the phantom. Discrepancies for central inserts are caused by beam hardening issues. The root mean square deviation in the SPR for the best SECT (90kV/Cu) and DECT (50kV/Al-90kV/Al) protocols are 3.7% and 1.0%, respectively. Dose distributions calculated for SECT and DECT datasets revealed range shifts <0.1 mm, gamma pass rates (3%/0.1mm) greater than 99%, and no substantial dosimetric differences for all structures. The outcomes suggest that SECT is sufficient for proton treatment planning in animals.<br><b>Significance:</b> The framework is a useful tool for the development of an optimized experimental configuration without using animals and beam time.},
}
@article{Jia2024b,
abbr = {},
bibtex_show = {true},
title = {Explainable fully automated CT scoring of interstitial lung disease for patients suspected of systemic sclerosis by cascaded regression neural networks and its comparison with experts},
author = {Jia, Jingnan and Hern{\'a}ndez Gir{\'o}n, Irene and Schouffoer, Anne A. and De Vries-Bouwstra, Jeska K. and Ninaber, Maarten K. and Korving, Julie C. and Staring, Marius and Kroft, Lucia J.M. and Stoel, Berend C.},
journal = {Scientific Reports},
volume = {14},
pages = {26666},
month = {},
year = {2024},
pdf = {2024_j_SR.pdf},
html = {https://doi.org/10.1038/s41598-024-78393-4},
arxiv = {},
code = {},
abstract = {Visual scoring of interstitial lung disease in systemic sclerosis (SSc-ILD) from CT scans is laborious, subjective and time-consuming. This study aims to develop a deep learning framework to automate SSc-ILD scoring. The automated framework is a cascade of two neural networks. The first network selects the craniocaudal positions of the five scoring levels. Subsequently, for each level, the second network estimates the ratio of three patterns to the total lung area: the total extent of disease (TOT), ground glass (GG) and reticulation (RET). To overcome the score imbalance in the second network, we propose a method to augment the training dataset with synthetic data. To explain the network's output, a heat map method is introduced to highlight the candidate interstitial lung disease regions. The explainability of heat maps was evaluated by two human experts and a quantitative method that uses the heat map to produce the score. The results show that our framework achieved a κ of 0.66, 0.58, and 0.65, for the TOT, GG and RET scoring, respectively. Both experts agreed with the heat maps in 91%, 90% and 80% of cases, respectively. Therefore, it is feasible to develop a framework for automated SSc-ILD scoring, which performs competitively with human experts and provides high-quality explanations using heat maps. Confirming the model's generalizability is needed in future studies.},
}
@article{Jia2024a,
abbr = {},
bibtex_show = {true},
title = {Using 3D point cloud and graph-based neural networks to improve the estimation of pulmonary function tests from chest CT},
author = {Jia, Jingnan and Yu, Bo and Mody, Prerak and Ninaber, Maarten K. and Schouffoer, Anne A. and Kroft, Lucia J.M. and Staring, Marius and Stoel, Berend C.},
journal = {Computers in Biology and Medicine},
volume = {182},
pages = {109192},
month = {November},
year = {2024},
pdf = {2024_j_CMB.pdf},
html = {https://doi.org/10.1016/j.compbiomed.2024.109192},
arxiv = {},
code = {https://github.com/Jingnan-Jia/PFT_regression},
abstract = {Pulmonary function tests (PFTs) are important clinical metrics to measure the severity of interstitial lung disease for systemic sclerosis patients. However, PFTs cannot always be performed by spirometry if there is a risk of disease transmission or other contraindications. In addition, it is unclear how lung function is affected by changes in lung vessels. Convolution neural networks (CNNs) have been previously proposed to estimate PFTs from chest CT scans (CNN-CT) and extracted vessels (CNNVessel). Due to GPU memory constraints, however, these networks used down-sampled images, which causes a loss of information on small vessels. Previous work based on CNNs has indicated that detailed vessel information from CT scans can be helpful for PFT estimation. Therefore, this paper proposes to use a point cloud neural network (PNN-Vessel) and graph neural network (GNN-Vessel) to estimate PFTs from point cloud and graph-based representations of pulmonary vessel centerlines, respectively. After that, we perform multiple variable step-wise regression analysis to explore if vessel-based networks can contribute to the PFT estimation, in addition to CNN-CT. Results showed that both PNN-Vessel and GNN-Vessel outperformed CNN-Vessel, by 14% and 4%, respectively, when averaged across the ICC scores of four PFTs metrics. In addition, compared to CNN-Vessel, PNNVessel used 30% of training time (1.1 hours) and 7% parameters (2.1 M) and GNN-Vessel used only 7% training time (0.25 hours) and 0.7% parameters (0.2 M). Our multiple variable regression analysis still verified that more detailed vessel information could provide further explanation for PFT estimation from anatomical imaging.},
}
@article{Mody:2024b,
abbr = {},
bibtex_show = {true},
title = {Improving Uncertainty-Error Correspondence in Deep Bayesian Medical Image Segmentation},
author = {Mody, Prerak and Chaves-de-Plaza, Nicolas and Rao, Chinmay and Astreinidou, Eleftheria and De Ridder, Mischa, and Hoekstra, Nienke and Hildebrandt, Klaus and Staring, Marius},
journal = {The Journal of Machine Learning for Biomedical Imaging},
volume = {2},
pages = {1048 -- 1082},
month = {August},
year = {2024},
pdf = {2024_j_MELBAb.pdf},
html = {https://doi.org/10.59275/j.melba.2024-5gc8},
arxiv = {},
code = {https://github.com/prerakmody/bayesuncertainty-error-correspondence},
abstract = {Increased usage of automated tools like deep learning in medical image segmentation has alleviated the bottleneck of manual contouring. This has shifted manual labour to quality assessment (QA) of automated contours which involves detecting errors and correcting them. A potential solution to semi-automated QA is to use deep Bayesian uncertainty to recommend potentially erroneous regions, thus reducing time spent on error detection. Previous work has investigated the correspondence between uncertainty and error, however, no work has been done on improving the ``utility" of Bayesian uncertainty maps such that it is only present in inaccurate regions and not in the accurate ones. Our work trains the FlipOut model with the Accuracy-vs-Uncertainty (AvU) loss which promotes uncertainty to be present only in inaccurate regions. We apply this method on datasets of two radiotherapy body sites, c.f. head-and-neck CT and prostate MR scans. Uncertainty heatmaps (i.e. predictive entropy) are evaluated against voxel inaccuracies using Receiver Operating Characteristic (ROC) and Precision-Recall (PR) curves. Numerical results show that when compared to the Bayesian baseline the proposed method successfully suppresses uncertainty for accurate voxels, with similar presence of uncertainty for inaccurate voxels. Code to reproduce experiments is available at <a href="https://github.com/prerakmody/bayesuncertainty-error-correspondence">https://github.com/prerakmody/bayesuncertainty-error-correspondence</a>.}
}
@article{Chaves-de-Plaza:2024,
abbr = {CGF},
bibtex_show = {true},
title = {Depth for Multi-Modal Contour Ensembles},
author = {Chaves-de-Plaza, N.F. and Molenaar, M. and Mody, P. and Staring, M. and van Egmond, R. and Eisemann, E. and Vilanova, A. and Hildebrandt, K.},
journal = {Computer Graphics Forum},
volume = {43},
number = {3},
pages = {e15083},
year = {2024},
pdf = {2024_j_CGF.pdf},
html = {https://doi.org/10.1111/cgf.15083},
code = {https://github.com/chadepl/paper-multimodal-contour-depth},
abstract = {The contour depth methodology enables non-parametric summarization of contour ensembles by extracting their representatives, confidence bands, and outliers for visualization (via contour boxplots) and robust downstream procedures. We address two shortcomings of these methods. Firstly, we significantly expedite the computation and recomputation of Inclusion Depth (ID), introducing a linear-time algorithm for epsilon ID, a variant used for handling ensembles with contours with multiple intersections. We also present the inclusion matrix, which contains the pairwise inclusion relationships between contours, and leverage it to accelerate the recomputation of ID. Secondly, extending beyond the single distribution assumption, we present the Relative Depth (ReD), a generalization of contour depth for ensembles with multiple modes. Building upon the linear-time eID, we introduce CDclust, a clustering algorithm that untangles ensemble modes of variation by optimizing ReD. Synthetic and real datasets from medical image segmentation and meteorological forecasting showcase the speed advantages, illustrate the use case of progressive depth computation and enable non-parametric multimodal analysis. To promote research and adoption, we offer the contour-depth Python package.},
}
@article{Mody:2024a,
abbr = {PhIRO},
bibtex_show = {true},
title = {Large-scale dose evaluation of deep learning organ contours in head-and-neck radiotherapy by leveraging existing plans},
author = {Mody, Prerak and Huiskes, Merle and Chaves-de-Plaza, Nicolas and Onderwater, Alice and Lamsma, Rense and Hildebrandt, Klaus and Hoekstra, Nienke and Astreinidou, Eleftheria and Staring, Marius and Dankers, Frank},
journal = {Physics and Imaging in Radiation Oncology},
volume = {30},
pages = {100572},
month = {April},
year = {2024},
pdf = {2024_j_PHIRO.pdf},
html = {https://doi.org/10.1016/j.phro.2024.100572},
code = {https://github.com/prerakmody/dose-eval-via-existing-plan-parameters},
abstract = {<b>Background and Purpose:</b> Retrospective dose evaluation for organ-at-risk auto-contours has previously used small cohorts due to additional manual effort required for treatment planning on auto-contours. We aimed to do this at large scale, by a) proposing and assessing an automated plan optimization workflow that used existing clinical plan parameters and b) using it for head-and-neck auto-contour dose evaluation.<br><b>Materials and Methods:</b> Our automated workflow emulated our clinic's treatment planning protocol and reused existing clinical plan optimization parameters. This workflow recreated the original clinical plan (P<sub>OG</sub>) with manual contours (P<sub>MC</sub>) and evaluated the dose effect (P<sub>OG</sub> - P<sub>MC</sub>) on 70 photon and 30 proton plans of head-and-neck patients. As a use-case, the same workflow (and parameters) created a plan using auto-contours (P<sub>AC</sub>) of eight head-and-neck organs-at-risk from a commercial tool and evaluated their dose effect (P<sub>MC</sub> - P<sub>AC</sub>).<br><b>Results:</b> For plan recreation (P<sub>OG</sub> - P<sub>MC</sub>), our workflow had a median impact of 1.0% and 1.5% across dose metrics of auto-contours, for photon and proton respectively. Computer time of automated planning was 25% (photon) and 42% (proton) of manual planning time. For auto-contour evaluation (P<sub>MC</sub> - P<sub>AC</sub>), we noticed an impact of 2.0% and 2.6% for photon and proton radiotherapy. All evaluations had a median ΔNTCP (Normal Tissue Complication Probability) less than 0.3%.<br><b>Conclusions:</b> The plan replication capability of our automated program provides a blueprint for other clinics to perform auto-contour dose evaluation with large patient cohorts. Finally, despite geometric differences, auto-contours had a minimal median dose impact, hence inspiring confidence in their utility and facilitating their clinical adoption.}
}
@article{Stoel:2024,
abbr = {Nat. Rev. Rheumatol.},
bibtex_show = {true},
title = {Deep Learning in Rheumatologic Image Interpretation},
author = {Stoel, Berend C. and Staring, Marius and Reijnierse, Monique and van der Helm-van Mil, Annette H.M.},
journal = {Nature Reviews Rheumatology},
volume = {20},
pages = {182 -- 195},
month = {March},
year = {2024},
pdf = {2024_j_NRR.pdf},
html = {https://doi.org/10.1038/s41584-023-01074-5},
abstract = {Artificial intelligence techniques, specifically deep learning, have already affected daily life in a wide range of areas. Likewise, initial applications have been explored in rheumatology. Deep learning might not easily surpass the accuracy of classic techniques when performing classification or regression on low-dimensional numerical data. With images as input, however, deep learning has become so successful that it has already outperformed the majority of conventional image-processing techniques developed during the past 50 years. As with any new imaging technology, rheumatologists and radiologists need to consider adapting their arsenal of diagnostic, prognostic and monitoring tools, and even their clinical role and collaborations. This adaptation requires a basic understanding of the technical background of deep learning, to efficiently utilize its benefits but also to recognize its drawbacks and pitfalls, as blindly relying on deep learning might be at odds with its capabilities. To facilitate such an understanding, it is necessary to provide an overview of deep-learning techniques for automatic image analysis in detecting, quantifying, predicting and monitoring rheumatic diseases, and of currently published deep-learning applications in radiological imaging for rheumatology, with critical assessment of possible limitations, errors and confounders, and conceivable consequences for rheumatologists and radiologists in clinical practice.}
}
@article{Chen:2024,
abbr = {Melba},
bibtex_show = {true},
author = {Chen, Yunjie and Staring, Marius and Neve, Olaf M. and Romeijn, Stephan R. and Hensen, Erik F. and Verbist, Berit M. and Wolterink, Jelmer M. and Tao, Qian},
title = {CoNeS: Conditional neural fields with shift modulation for multi-sequence MRI translation},
journal = {The Journal of Machine Learning for Biomedical Imaging},
volume = {2},
pages = {657 -- 685},
year = {2024},
pdf = {2024_j_MELBAa.pdf},
html = {https://doi.org/10.59275/j.melba.2024-d61g},
arxiv = {2309.03320},
code = {https://github.com/cyjdswx/CoNeS.git},
abstract = {Multi-sequence magnetic resonance imaging (MRI) has found wide applications in both modern clinical studies and deep learning research. However, in clinical practice, it frequently occurs that one or more of the MRI sequences are missing due to different image acquisition protocols or contrast agent contraindications of patients, limiting the utilization of deep learning models trained on multi-sequence data. One promising approach is to leverage generative models to synthesize the missing sequences, which can serve as a surrogate acquisition. State-of-the-art methods tackling this problem are based on convolutional neural networks (CNN) which usually suffer from spectral biases, resulting in poor reconstruction of high-frequency fine details. In this paper, we propose Conditional Neural fields with Shift modulation (CoNeS), a model that takes voxel coordinates as input and learns a representation of the target images for multi-sequence MRI translation. The proposed model uses a multi-layer perceptron (MLP) instead of a CNN as the decoder for pixel-to-pixel mapping. Hence, each target image is represented as a neural field that is conditioned on the source image via shift modulation with a learned latent code. Experiments on BraTS 2018 and an in-house clinical dataset of vestibular schwannoma patients showed that the proposed method outperformed state-of-the-art methods for multi-sequence MRI translation both visually and quantitatively. Moreover, we conducted spectral analysis, showing that CoNeS was able to overcome the spectral bias issue common in conventional CNN models. To further evaluate the usage of synthesized images in clinical downstream tasks, we tested a segmentation network using the synthesized images at inference. The results showed that CoNeS improved the segmentation performance when some MRI sequences were missing and outperformed other synthesis models. We concluded that neural fields are a promising technique for multi-sequence MRI translation.},
}
@article{Chaves-de-Plaza:2024,
abbr = {TVCG},
bibtex_show = {true},
author = {Chaves-de-Plaza, Nicolas and Mody, Prerak P. and Staring, Marius and van Egmond, Ren{\'e}; and Vilanova, Anna and Hildebrandt, Klaus},
title = {Inclusion Depth for Contour Ensembles},
journal = {IEEE Transactions on Visualization and Computer Graphics},
volume = {},
number = {},
pages = {},
year = {2024},
pdf = {2024_j_TVCG.pdf},
html = {https://doi.org/10.1109/TVCG.2024.3350076},
arxiv = {},
code = {},
abstract = {Ensembles of contours arise in various applications like simulation, computer-aided design, and semantic segmentation. Uncovering ensemble patterns and analyzing individual members is a challenging task that suffers from clutter. Ensemble statistical summarization can alleviate this issue by permitting analyzing ensembles' distributional components like the mean and median, confidence intervals, and outliers. Contour boxplots, powered by Contour Band Depth (CBD), are a popular nonparametric ensemble summarization method that benefits from CBD's generality, robustness, and theoretical properties. In this work, we introduce Inclusion Depth (ID), a new notion of contour depth with three defining characteristics. First, ID is a generalization of functional Half-Region Depth, which offers several theoretical guarantees. Second, ID relies on a simple principle: the inside/outside relationships between contours. This facilitates implementing ID and understanding its results. Third, the computational complexity of ID scales quadratically in the number of members of the ensemble, improving CBD's cubic complexity. This also in practice speeds up the computation enabling the use of ID for exploring large contour ensembles or in contexts requiring multiple depth evaluations like clustering. In a series of experiments on synthetic data and case studies with meteorological and segmentation data, we evaluate ID's performance and demonstrate its capabilities for the visual analysis of contour ensembles.},
}
@article{Beljaards:2024,
abbr = {},
bibtex_show = {true},
author = {Beljaards, Laurens and Pezzotti, Nicola and Rao, Chinmay and Doneva, Mariya and van Osch, Matthias J.P. and Staring, Marius},
title = {AI-Based Motion Artifact Severity Estimation in Undersampled MRI Allowing for Selection of Appropriate Reconstruction Models},
journal = {Medical Physics},
volume = {51},
number = {5},
pages = {3555 -- 3565},
year = {2024},
pdf = {2024_j_MP.pdf},
html = {https://doi.org/10.1002/mp.16918},
arxiv = {},
code = {},
abstract = {<b>Background:</b> MR acquisition is a time consuming process, making it susceptible to patient motion during scanning. Even motion in the order of a millimeter can introduce severe blurring and ghosting artifacts, potentially necessitating re-acquisition. MRI can be accelerated by acquiring only a fraction of k-space, combined with advanced reconstruction techniques leveraging coil sensitivity profiles and prior knowledge. AI-based reconstruction techniques have recently been popularized, but generally assume an ideal setting without intra-scan motion.<br><b>Purpose:</b> To retrospectively detect and quantify the severity of motion artifacts in undersampled MRI data. This may prove valuable as a safety mechanism for AI-based approaches, provide useful information to the reconstruction method, or prompt for re-acquisition while the patient is still in the scanner.<br><b>Methods:</b> We developed a deep learning approach that detects and quantifies motion artifacts in undersampled brain MRI. We demonstrate that synthetically motion-corrupted data can be leveraged to train the CNN-based motion artifact estimator, generalizing well to real-world data. Additionally, we leverage the motion artifact estimator by using it as a selector for a motion-robust reconstruction model in case a considerable amount of motion was detected, and a high data consistency model otherwise.<br><b>Results:</b> Training and validation were performed on 4387 and 1304 synthetically motion-corrupted images and their uncorrupted counterparts, respectively. Testing was performed on undersampled in vivo motion-corrupted data from 28 volunteers, where our model distinguished head motion from motion-free scans with 91% and 96% accuracy when trained on synthetic and on real data, respectively. It predicted a manually defined quality label (`Good', `Medium' or `Bad' quality) correctly in 76% and 85% of the time when trained on synthetic and real data, respectively. When used as a selector it selected the appropriate reconstruction network 93% of the time, achieving near optimal SSIM values.<br><b>Conclusions:</b> The proposed method quantified motion artifact severity in undersampled MRI data with high accuracy, enabling real-time motion artifact detection that can help improve the safety and quality of AI-based reconstructions.},
}
@article{Jia:2023,
abbr = {},
bibtex_show = {true},
author = {Jia, Jingnan and Marges, Emiel R. and Ninaber, Maarten K. and Kroft, Lucia J.M. and Schouffoer, Anne A. and Staring, Marius and Stoel, Berend C.},
title = {Automatic pulmonary function estimation from chest CT scans using deep regression neural networks: the relation between structure and function in systemic sclerosis},
journal = {IEEE Access},
volume = {11},
pages = {135272 -- 135282},
month = {November},
year = {2023},
pdf = {2023_j_Access.pdf},
html = {https://doi.org/10.1109/ACCESS.2023.3337639},
arxiv = {},
code = {},
abstract = {Pulmonary function test (PFT) plays an important role in screening and following-up pulmonary involvement in systemic sclerosis (SSc). However, some patients are not able to perform PFT due to contraindications. In addition, it is unclear how lung function is affected by changes in lung structure in SSc. Therefore, this study aims to explore the potential of automatically estimating PFT results from chest CT scans of SSc patients and how different regions influence the estimation of PFT values. Deep regression networks were developed with transfer learning to estimate PFT from 316 SSc patients. Segmented lungs and vessels were used to mask the CT images to train the network with different inputs: from entire CT scan, lungs-only to vessels-only. The network trained by entire CT scans with transfer learning achieved an ICC of 0.71, 0.76, 0.80, and 0.81 for the estimation of DLCO, FEV1, FVC and TLC, respectively. The performance of the networks gradually decreased when trained on data from lungs-only and vessels-only. Regression attention maps showed that regions close to large vessels are highlighted more than other regions, and occasionally regions outside the lungs are highlighted. These experiments mean that apart from lungs and large vessels, other regions contribute to the estimation of PFTs. In addition, adding manually designed biomarkers increased the correlation (R) from 0.75, 0.74, 0.82, and 0.83 to 0.81, 0.83, 0.88, and 0.90, respectively. It means that that manually designed imaging biomarkers can still contribute to explaining the relation between lung function and structure.},
}
@article{Neve:2023,
abbr = {},
bibtex_show = {true},
author = {Neve, Olaf M. and Romeijn, Stephan R. and Chen, Yunjie and Nagtegaal, Larissa and Grootjans, Willem and Jansen, Jeroen C. and Staring, Marius and Verbist, Berit M. and Hensen, Erik F.},
title = {Automated 2-dimensional measurement of vestibular schwannoma: validity and accuracy of an artificial intelligence algorithm},
journal = {Otolaryngology - Head and Neck Surgery},
volume = {169},
number = {6},
pages = {1582 -- 1589},
month = {December},
year = {2023},
pdf = {2023_j_OHNS.pdf},
html = {https://doi.org/10.1002/ohn.470},
arxiv = {},
code = {},
abstract = {<b>Objective.</b> Validation of automated 2-dimensional (2D) diameter measurements of vestibular schwannomas on magnetic resonance imaging (MRI).<br><b>Study Design.</b>Retrospective validation study using 2 data sets containing MRIs of vestibular schwannoma patients.<br><b>Setting.</b> University Hospital in The Netherlands.<br><b>Methods.</b>Two data sets were used, 1 containing 1 scan per patient (n = 134) and the other containing at least 3 consecutive MRIs of 51 patients, all with contrast-enhanced T1 or high-resolution T2 sequences. 2D measurements of the maximal extrameatal diameters in the axial plane were automatically derived from a 3D-convolutional neural network compared to manual measurements by 2 human observers. Intra- and interobserver variabilities were calculated using the intraclass correlation coefficient (ICC), agreement on tumor progression using Cohen's kappa.<br><b>Results.</b> The human intra- and interobserver variability showed a high correlation (ICC: 0.98-0.99) and limits of agreement of 1.7 to 2.1 mm. Comparing the automated to human measurements resulted in ICC of 0.98 (95% confidence interval [CI]: 0.974; 0.987) and 0.97 (95% CI: 0.968; 0.984), with limits of agreement of 2.2 and 2.1 mm for diameters parallel and perpendicular to the posterior side of the temporal bone, respectively. There was satisfactory agreement on tumor progression between automated measurements and human observers (Cohen's κ = 0.77), better than the agreement between the human observers (Cohen's κ = 0.74).<br><b>Conclusion.</b> Automated 2D diameter measurements and growth detection of vestibular schwannomas are at least as accurate as human 2D measurements. In clinical practice, measurements of the maximal extrameatal tumor (2D) diameters of vestibular schwannomas provide important complementary information to total tumor volume (3D) measurements. Combining both in an automated measurement algorithm facilitates clinical adoption.},
}
@article{Zhai:2023,
abbr = {},
bibtex_show = {true},
author = {Zhai, Zhiwei and Boon, Gudula J.A.M. and Staring, Marius and van Dam, Lisette F. and Kroft, Lucia J.M. and Giron, Irene Hernandez and Ninaber, Maarten K. and Bogaard, Harm Jan and Meijboom, Lilian J. and Vonk Noordegraaf, Anton and Huisman, Menno V. and Klok, Frederikus A. and Stoel, Berend C.},
title= {Automated Quantification of the Pulmonary Vasculature in Pulmonary Embolism and Chronic Thromboembolic Pulmonary Hypertension},
journal = {Pulmonary Circulation},
volume = {13},
number = {2},
pages = {e12223},
year = {2023},
pdf = {2023_j_PC.pdf},
html = {https://doi.org/10.1002/pul2.12223},
arxiv = {},
code = {},
abstract = {The particular mechanical obstruction of pulmonary embolism (PE) and chronic thromboembolic pulmonary hypertension (CTEPH) may affect pulmonary arteries and veins differently. Therefore, we evaluated whether pulmonary vascular morphology and densitometry using CT pulmonary angiography (CTPA) in arteries and veins could distinguish PE from CTEPH.<br>We analyzed CTPA images from a convenience cohort of 16 PE patients, 6 CTEPH patients and 15 controls without PE or CTEPH. Pulmonary vessels were extracted with a graph-cuts method, and separated into arteries and veins using a deep-learning classification method. By analyzing the distribution of vessel radii, vascular morphology was quantified into a slope (α) and intercept (β) for the entire pulmonary vascular tree, and for arteries and veins, separately. To quantify lung perfusion, the median pulmonary vascular density was calculated. As a reference, lung perfusion was also quantified by the contrast enhancement in the parenchymal areas, pulmonary trunk and descending aorta. All quantifications were compared between the three groups.<br>Vascular morphology did not differ between groups, in contrast to vascular density values (both arterial and venous; p-values 0.006 - 0.014). The median vascular density (interquartile range) was -452 (95), -567 (113) and -470 (323) HU, for the PE, control and CTEPH group, respectively. The perfusion curves from all measurements showed different patterns between groups.<br>In this proof of concept study, not vasculature morphology but vascular densities differentiated between normal and thrombotic obstructed vasculature. For distinction on an individual patient level, further technical improvements are needed both in terms of image acquisition/reconstruction and post-processing.},
}
@article{Goedmakers:2022,
abbr = {},
bibtex_show = {true},
author = {Goedmakers, C.M.W and Pereboom, L.M. and Schoones, J.W. and de Leeuw den Bouter, M.L. and Remis, R.F. and Staring, M. and Vleggeert-Lankamp, C.L.A.},
title = {Machine learning for image analysis in the cervical spine: Systematic review of the available models and methods},
journal = {Brain and Spine},
volume = {2},
pages = {101666},
year = {2022},
pdf = {2022_j_BandS.pdf},
html = {https://doi.org/10.1016/j.bas.2022.101666},
arxiv = {},
code = {},
abstract = {<ul><li>Neural network approaches show the most potential for automated image analysis of thecervical spine.</li><li>Fully automatic convolutional neural network (CNN) models are promising Deep Learning methods for segmentation.</li><li>In cervical spine analysis, the biomechanical features are most often studied using finiteelement models.</li><li>The application of artificial neural networks and support vector machine models looks promising for classification purposes.</li><li>This article provides an overview of the methods for research on computer aided imaging diagnostics of the cervical spine.</li></ul>},
}
@article{Neve:2022,
abbr = {Radiol Artif Intell},
bibtex_show = {true},
author = {Neve, Olaf and Chen, Yunjie and Tao, Qian and Romeijn, Stephan and de Boer, Nick and Grootjans, Willem and Kruit, Mark and Lelieveldt, Boudewijn and Jansen, Jeroen and Hensen, Erik and Verbist, Berit and Staring, Marius},
title = {Fully Automated 3D Vestibular Schwannoma Segmentation with and without Gadolinium Contrast: a multi-center, multi-vendor study},
journal = {Radiology: Artificial Intelligence},
volume = {4},
number = {4},
pages = {e210300},
year = {2022},
pdf = {2022_j_RadAI.pdf},
html = {https://doi.org/10.1148/ryai.210300},
abstract = {<b>Purpose:</b> To develop automated vestibular schwannoma measurements on contrast-enhanced T1- and T2-weighted MRI.<br><b>Material and methods:</b> MRI data from 214 patients in 37 different centers was retrospectively analyzed between 2020-2021. Patients with hearing loss (134 vestibular schwannoma positive [mean age ± SD, 54 ± 12 years; 64 men], 80 negative) were randomized to a training and validation set and an independent test set. A convolutional neural network (CNN) was trained using five-fold cross-validation for two models (T1 and T2). Quantitative analysis including Dice index, Hausdorff distance, surface-to-surface distance (S2S), and relative volume error were used to compare the computer and the human delineations. Furthermore, an observer study was performed in which two experienced physicians evaluated both delineations.<br><b>Results:</b> The T1-weighted model showed state-of-the-art performance with a mean S2S distance of less than 0.6 mm for the whole tumor and the intrameatal and extrameatal tumor parts. The whole tumor Dice index and Hausdorff distance were 0.92 and 2.1 mm in the independent test set. T2-weighted images had a mean S2S distance less than 0.6 mm for the whole tumor and the intrameatal and extrameatal tumor parts. Whole tumor Dice index and Hausdorff distance were 0.87 and 1.5 mm in the independent test set. The observer study indicated that the tool was comparable to human delineations in 85-92% of cases.<br><b>Conclusion:</b> The CNN model detected and delineated vestibular schwannomas accurately on contrast-enhanced T1 and T2-weighted MRI and distinguished the clinically relevant difference between intrameatal and extrameatal tumor parts.}
}
@article{Koolstra:2022,
abbr = {},
bibtex_show = {true},
author = {Koolstra, Kirsten and Staring, Marius and de Bruin, Paul and van Osch, Mathias J.P.},
title = {Subject-specific optimization of background suppression for arterial spin labeling MRI using a feedback loop on the scanner},
journal = {NMR in Biomedicine},
volume = {35},
number = {9},
pages = {e4746},
month = {September},
year = {2022},
pdf = {2022_j_NMR.pdf},
html = {https://doi.org/10.1002/nbm.4746},
arxiv = {},
code = {},
abstract = {Background suppression (BGS) in arterial spin labeling (ASL) MRI leads to a higher temporal SNR (tSNR) of the perfusion images compared to ASL without BGS. The performance of the BGS, however, depends on the tissue relaxation times and on inhomogeneities of the scanner's magnetic fields, which differ between subjects and are unknown at the moment of scanning. Therefore, we developed a feedback loop (FBL) mechanism that optimizes the BGS for each subject in the scanner during acquisition. We implemented the FBL for 2D pseudo-continuous ASL (PCASL) scans with an echo-planar imaging (EPI) readout. After each dynamic scan, acquired ASL images were automatically sent to an external computer and processed with a Python processing tool. Inversion times were optimized on-the-fly using 80 iterations of the Nelder-Mead method, by minimizing the signal intensity in the label image while maximizing the signal intensity in the perfusion image. The performance of this method was first tested in a 4-component phantom. The regularization parameter was then tuned in 6 healthy subjects (3 male, 3 female, age 24-62 years) and set as λ=4 for all other experiments. Resulting ASL images, perfusion images and tSNR maps obtained from the last 20 iterations of the FBL scan were compared to those obtained without BGS and to standard BGS in 12 healthy volunteers (5 male, 7 female, age 24-62 years) (including the 6 volunteers used for tuning of λ). The FBL resulted in perfusion images with a statistically significantly higher tSNR (2.20) compared to standard BGS (1.96) (P < 5 10<sup>-3</sup>, two-sided paired t-test). Minimizing signal in the label image furthermore resulted in control images from which approximate changes in perfusion signal can directly be appreciated. This could be relevant to ASL applications that require a high temporal resolution. Future work is needed to minimize the number of initial acquisitions during which the performance of BGS is reduced compared to standard BGS and to extend the technique to 3D ASL.},
}
@article{Brink:2022,
abbr = {},
bibtex_show = {true},
author = {Brink, Wyger M. and Yousefi, Sahar and Bhatnagar, Prerna and Remis, Rob F. and Staring, Marius and Webb, Andrew G.},
title = {Personalised Local SAR Prediction for Parallel Transmit Neuroimaging at 7T from a Single T1-weighted Dataset},
journal = {Magnetic Resonance in Medicine},
volume = {88},
number = {1},
pages = {464 - 475},
month = {July},
year = {2022},
pdf = {2022_j_MRM.pdf},
html = {https://doi.org/10.1002/mrm.29215},
arxiv = {},
code = {https://github.com/wygerbrink/PersonalizedDosimetry},
abstract = {<b>Purpose.</b> Parallel RF transmission (PTx) is one of the key technologies enabling high quality imaging at ultrahigh field strengths (≥7T). Compliance with regulatory limits on the local specific absorption rate (SAR) typically involves over-conservative safety margins to account for intersubject variability, which negatively affect the utilization of ultra-high field MR. In this work, we present a method to generate a subject-specific body model from a single T1-weighted dataset for personalized local SAR prediction in PTx neuroimaging at 7T.<br><b>Methods.</b> Multi-contrast data were acquired at 7T (N=10) to establish ground truth segmentations in eight tissue types. A 2.5D convolutional neural network was trained using the T1-weighted data as input in a leave-one-out cross-validation study. The segmentation accuracy was evaluated through local SAR simulations in a quadrature birdcage as well as a PTx coil model.<br><b>Results.</b> The network-generated segmentations reached overall Dice coefficients of 86.7% ± 6.7% (mean ± standard deviation) and showed to successfully address the severe intensity bias and contrast variations typical to 7T. Errors in peak local SAR obtained were below 3.0% in the quadrature birdcage. Results obtained in the PTx configuration indicated that a safety margin of 6.3% ensures conservative local SAR estimates in 95% of the random RF shims, compared to an average overestimation of 34% in the generic "one-size-fits-all" approach.<br><b>Conclusion.</b> A subject-specific body model can be automatically generated from a single T1-weighted dataset by means of deep learning, providing the necessary inputs for accurate and personalized local SAR predictions in PTx neuroimaging at 7T.},
}
@article{Malimban:2022,
abbr = {},
bibtex_show = {true},
author = {Malimban, Justin and Lathouwers, Danny and Qian, Haibin and Verhaegen, Frank and Wiedemann, Julia and Brandenburg, Sytze and Staring, Marius},
title = {Deep learning-based segmentation of the thorax in mouse micro-CT scans},
journal = {Scientific Reports},
volume = {12},
number = {1},
pages = {1822},
year = {2022},
pdf = {2022_j_SR.pdf},
html = {https://doi.org/10.1038/s41598-022-05868-7},
arxiv = {},
code = {},
abstract = {For image-guided small animal irradiations, the whole workflow of imaging, organ contouring, irradiation planning, and delivery is typically performed in a single session requiring continuous administration of anesthetic agents. Automating contouring leads to a faster workflow, which limits exposure to anesthesia and thereby, reducing its impact on experimental results and on animal wellbeing. Here, we trained the 2D and 3D U-Net architectures of no-new-Net (nnU-Net) for autocontouring of the thorax in mouse micro-CT images. We trained the models only on native CTs and evaluated their performance using an independent testing dataset (i.e., native CTs not included in the training and validation). Unlike previous studies, we also tested the model performance on an external dataset (i.e., contrast-enhanced CTs) to see how well they predict on CTs completely different from what they were trained on. We also assessed the interobserver variability using the generalized conformity index (CIgen) among three observers, providing a stronger human baseline for evaluating automated contours than previous studies. Lastly, we showed the benefit on the contouring time compared to manual contouring. The results show that 3D models of nnU-Net achieve superior segmentation accuracy and are more robust to unseen data than 2D models. For all target organs, the mean surface distance (MSD) and the Hausdorff distance (95p HD) of the best performing model for this task (nnU-Net 3d_fullres) are within 0.16 mm and 0.60 mm, respectively. These values are below the minimum required contouring accuracy of 1 mm for small animal irradiations, and improve significantly upon state-of-the-art 2D U-Net-based AIMOS method. Moreover, the conformity indices of the 3d_fullres model also compare favourably to the interobserver variability for all target organs, whereas the 2D models perform poorly in this regard. Importantly, the 3d_fullres model offers 98% reduction in contouring time.},
}
@article{Elmahdy:2021,
abbr = {},
bibtex_show = {true},
author = {Elmahdy, Mohamed S. and Beljaards, Laurens and Yousefi, Sahar and Sokooti, Hessam and Verbeek, Fons and van der Heide, U.A. and Staring, Marius},
title = {Joint Registration and Segmentation via Multi-Task Learning for Adaptive Radiotherapy of Prostate Cancer},
journal = {IEEE Access},
volume = {9},
pages = {95551 -- 95568},
month = {June},
year = {2021},
pdf = {2021_j_Accessc.pdf},
html = {https://doi.org/10.1109/ACCESS.2021.3091011},
arxiv = {2105.01844},
code = {https://github.com/moelmahdy/JRS-MTL},
abstract = {Medical image registration and segmentation are two of the most frequent tasks in medical image analysis. As these tasks are complementary and correlated, it would be beneficial to apply them simultaneously in a joint manner. In this paper, we formulate registration and segmentation as a joint problem via a Multi-Task Learning (MTL) setting, allowing these tasks to leverage their strengths and mitigate their weaknesses through the sharing of beneficial information. We propose to merge these tasks not only on the loss level, but on the architectural level as well. We studied this approach in the context of adaptive image-guided radiotherapy for prostate cancer, where planning and follow-up CT images as well as their corresponding contours are available for training. At testing time the contours of the follow-up scans are not available, which is a common scenario in adaptive radiotherapy. The study involves two datasets from different manufacturers and institutes. The first dataset was divided into training (12 patients) and validation (6 patients), and was used to optimize and validate the methodology, while the second dataset (14 patients) was used as an independent test set. We carried out an extensive quantitative comparison between the quality of the automatically generated contours from different network architectures as well as loss weighting methods. Moreover, we evaluated the quality of the generated deformation vector field (DVF). We show that MTL algorithms outperform their Single-Task Learning (STL) counterparts and achieve better generalization on the independent test set. The best algorithm achieved a mean surface distance of 1.06±0.3 mm, 1.27±0.4 mm, 0.91±0.4 mm, and 1.76±0.8 mm on the validation set for the prostate, seminal vesicles, bladder, and rectum, respectively. The high accuracy of the proposed method combined with the fast inference speed, makes it a promising method for automatic re-contouring of follow-up scans for adaptive radiotherapy, potentially reducing treatment related complications and therefore improving patients quality-of-life after treatment. The source code is available at <a href="https://github.com/moelmahdy/JRS-MTL">https://github.com/moelmahdy/JRS-MTL</a>.},
}
@article{Yousefi:2021,
abbr = {},
bibtex_show = {true},
author = {Yousefi, Sahar and Sokooti, Hessam and Elmahdy, Mohamed S. and Lips, Irene M. and Manzuri Shalmani, Mohammad T. and Zinkstok, Roel T. and Dankers, Frank J.W.M. and and Staring, Marius},
title = {Esophageal Tumor Segmentation in CT Images using a Dilated Dense Attention Unet (DDAUnet)},
journal = {IEEE Access},
volume = {9},
pages = {99235 -- 99248},
month = {July},
year = {2021},
pdf = {2021_j_Accessb.pdf},
html = {https://doi.org/10.1109/ACCESS.2021.3096270},
arxiv = {2012.03242},
code = {https://github.com/yousefis/DenseUnet_Esophagus_Segmentation},
abstract = {Manual or automatic delineation of the esophageal tumor in CT images is known to be very challenging. This is due to the low contrast between the tumor and adjacent tissues, the anatomical variation of the esophagus, as well as the occasional presence of foreign bodies (e.g. feeding tubes). Physicians therefore usually exploit additional knowledge such as endoscopic findings, clinical history, additional imaging modalities like PET scans. Achieving his additional information is time-consuming, while the results are error-prone and might lead to non-deterministic results. In this paper we aim to investigate if and to what extent a simplified clinical workflow based on CT alone, allows one to automatically segment the esophageal tumor with sufficient quality. For this purpose, we present a fully automatic end-to-end esophageal tumor segmentation method based on convolutional neural networks (CNNs). The proposed network, called Dilated Dense Attention Unet (DDAUnet), leverages spatial and channel attention gates in each dense block to selectively concentrate on determinant feature maps and regions. Dilated convolutional layers are used to manage GPU memory and increase the network receptive field. We collected a dataset of 792 scans from 288 distinct patients including varying anatomies with air pockets, feeding tubes and proximal tumors. Repeatability and reproducibility studies were conducted for three distinct splits of training and validation sets. The proposed network achieved a DSC value of 0.79 ± 0.20, a mean surface distance of 5.4 ± 20.2mm and 95% Hausdorff distance of 14.7 ± 25.0mm for 287 test scans, demonstrating promising results with a simplified clinical workflow based on CT alone. Our code is publicly available via <a href="https://github.com/yousefis/DenseUnet_Esophagus_Segmentation">https://github.com/yousefis/DenseUnet_Esophagus_Segmentation</a>.},
}
@article{Sokooti:2021,
abbr = {},
bibtex_show = {true},
author = {Sokooti, Hessam and Yousefi, Sahar and Elmahdy, Mohamed S. and Lelieveldt, Boudewijn P.F. and Staring, Marius},
title = {Hierarchical Prediction of Registration Misalignment using a Convolutional LSTM: Application to Chest CT Scans},
journal = {IEEE Access},
volume = {9},
pages = {62008 -- 62020},
month = {April},
year = {2021},
pdf = {2021_j_Accessa.pdf},
html = {https://doi.org/10.1109/ACCESS.2021.3074124},
arxiv = {},
code = {},
abstract = {In this paper we propose a supervised method to predict registration misalignment using convolutional neural networks (CNNs). This task is casted to a classification problem with multiple classes of misalignment: "correct" 0-3 mm, "poor" 3-6 mm and "wrong" over 6 mm. Rather than a direct prediction, we propose a hierarchical approach, where the prediction is gradually refined from coarse to fine. Our solution is based on a convolutional Long Short-Term Memory (LSTM), using hierarchical misalignment predictions on three resolutions of the image pair, leveraging the intrinsic strengths of an LSTM for this problem. The convolutional LSTM is trained on a set of artificially generated image pairs obtained from artificial displacement vector fields (DVFs). Results on chest CT scans show that incorporating multi-resolution information, and the hierarchical use via an LSTM for this, leads to overall better F1 scores, with fewer misclassifications in a well-tuned registration setup. The final system yields an accuracy of 87.1%, and an average F1 score of 66.4% aggregated in two independent chest CT scan studies.},
}
@article{Luu:2021,
abbr = {},
bibtex_show = {true},
author = {Luu, Ha Manh and van Walsum, Theo and Franklin, Daniel and Pham, Phuong Cam and Vu, Luu Dang and Moelker, Adriaan and Staring, Marius and Van Hoang, Xiem and Niessen, Wiro and Trung, Nguyen Linh},
title = {Efficiently Compressing 3D Medical Images for Teleinterventions via CNNs and Anisotropic Diffusion},
journal = {Medical Physics},
volume = {48},
number = {6},
pages = {2877 -- 2890},
month = {June},
year = {2021},
pdf = {2021_j_MP.pdf},
html = {https://doi.org/10.1002/mp.14814},
arxiv = {},
code = {},
abstract = {<b>Purpose:</b> Efficient compression of images while preserving image quality has the potential to be a major enabler of effective remote clinical diagnosis and treatment, since poor Internet connection conditions are often the primary constraint in such services. This paper presents a framework for organ-specific image compression for teleinterventions based on a deep learning approach and anisotropic diffusion filter.<br><b>Methods:</b> The proposed method, DLAD, uses a CNN architecture to extract a probability map for the organ of interest; this probability map guides an anisotropic diffusion filter that smooths the image except at the location of the organ of interest. Subsequently, a compression method, such as BZ2 and HEVC-visually lossless, is applied to compress the image. We demonstrate the proposed method on 3D CT images acquired for radio frequency ablation (RFA) of liver lesions. We quantitatively evaluate the proposed method on 151 CT images using peak-signal-to-noise ratio (PSNR), structural similarity (SSIM) and compression ratio (CR) metrics. Finally, we compare the assessments of two radiologists on the liver lesion detection and the liver lesion center annotation using 33 sets of the original images and the compressed images.<br><b>Results:</b> The results show that the method can significantly improve CR of most well-known compression methods. DLAD combined with HEVC-visually lossless achieves the highest average CR of 6.45, which is 36% higher than that of the original HEVC and outperforms other state-of-the-art lossless medical image compression methods. The means of PSNR and SSIM are 70 dB and 0.95, respectively. In addition, the compression effects do not statistically significantly affect the assessments of the radiologists on the liver lesion detection and the lesion center annotation.<br><b>Conclusions:</b> We thus conclude that the method has a high potential to be applied in teleintervention applications.},
}
@article{Pezzotti:2020,
abbr = {},
bibtex_show = {true},
author = {Pezzotti, Nicola and Yousefi, Sahar and Elmahdy, Mohamed S. and van Gemert, Jeroen and Sch{\"u}lke, Christophe and Doneva, Mariya and Nielsen, Tim and Kastryulin, Sergey and Lelieveldt, Boudewijn P.F. and van Osch, Matthias J.P. and de Weerdt, Elwin and Staring, Marius},
title = {An Adaptive Intelligence Algorithm for Undersampled Knee MRI Reconstruction},
journal = {IEEE Access},
volume = {8},
pages = {204825 -- 204838},
year = {2020},
pdf = {2020_j_Access.pdf},
html = {https://doi.org/10.1109/ACCESS.2020.3034287},
arxiv = {2004.07339},
code = {},
abstract = {Adaptive intelligence aims at empowering machine learning techniques with the additional use of domain knowledge. In this work, we present the application of adaptive intelligence to accelerate MR acquisition. Starting from undersampled k-space data, an iterative learning-based reconstruction scheme inspired by compressed sensing theory is used to reconstruct the images. We developed a novel deep neural network to refine and correct prior reconstruction assumptions given the training data. The network was trained and tested on a knee MRI dataset from the 2019 fastMRI challenge organized by Facebook AI Research and NYU Langone Health. All submissions to the challenge were initially ranked based on similarity with a known groundtruth, after which the top 4 submissions were evaluated radiologically. Our method was evaluated by the fastMRI organizers on an independent challenge dataset. It ranked #1, shared #1, and #3 on respectively the 8x accelerated multi-coil, the 4x multi-coil, and the 4x single-coil tracks. This demonstrates the superior performance and wide applicability of the method.},
}
@article{Zhao:2020,
abbr = {},
bibtex_show = {true},
author = {Zhao, Hong and Stoel, Berend C. and Staring, Marius and Bakker, M. Els and Stolk, Jan and Zhou, Ping and Xiao, Changyan},
title = {A framework for pulmonary fissure segmentation in 3D CT images using a directional derivative of plate filter},
journal = {Signal Processing},
volume = {173},
pages = {107602},
month = {August},
year = {2020},
pdf = {2020_j_SP.pdf},
html = {https://doi.org/10.1016/j.sigpro.2020.107602},
arxiv = {},
code = {},
abstract = {Imaging pulmonary fissures by CT provides useful information on diagnosis of pulmonary diseases. Automatic segmentation of fissures is a challenging task due to the variable appearance of fissures, such as inhomogeneous intensities, pathological deformation and imaging noise. To overcome these challenges, we propose an anisotropic differential operator called directional derivative of plate (DDoP) filter to probe the presence of fissure objects in 3D space by modeling the profile of a fissure patch with three parallel plates. To reduce the huge computation burden of dense matching with rotated DDoP kernels, a family of spherical harmonics are particularly utilized for acceleration. Additionally, a two-stage post-processing scheme is introduced to segment fissures. The performance of our method was verified in experiments using 55 scans from the publicly available LOLA11 dataset and 50 low-dose CT scans of lung cancer patients from the VIA-ELCAP database. Our method showed superior performance compared to the derivative of sticks (DoS) method and the Hessian-based method in terms of median and mean F1-score. The median F1-score for DDoP, DoS-based and Hessian-based methods on the LOLA11 dataset was 0.899, 0.848 and 0.843, respectively, and the mean F1-score was 0.858 ± 0.103, 0.781 ± 0.165 and 0.747 ± 0.239, respectively.},
}
@article{Yousefi:2019,
abbr = {},
bibtex_show = {true},
author = {Yousefi, Sahar and Manzuri Shalmani, M. T. and Lin, Jeremy and Staring, Marius},
title = {A Novel Motion Detection Method Using 3D Discrete Wavelet Transform},
journal = {IEEE Transactions on Circuits and Systems for Video Technology},
volume = {29},
number = {12},
pages = {3487 -- 3500},
month = {December},
year = {2019},
pdf = {2019_j_CSVT.pdf},
html = {https://doi.org/10.1109/TCSVT.2018.2885211},
arxiv = {},
code = {},
abstract = {The problem of motion detection has received considerable attention due to the explosive growth of its applications in video analysis and surveillance systems. While the previous approaches can produce good results, the accurate detection of motion remains a challenging task due to the difficulties raised by illumination variations, occlusion, camouflage, sudden motions appearing in burst, dynamic texture, and environmental changes such as those on weather conditions, sunlight changes during a day, etc. In this study, a novel per-pixel motion descriptor is proposed for motion detection in video sequences which outperforms the current methods in the literature particularly in severe scenarios. The proposed descriptor is based on two complementary three-dimensional discrete wavelet transforms (3D-DWT) and a three-dimensional wavelet leader. In this approach, a feature vector is extracted for each pixel by applying a novel three-dimensional wavelet-based motion descriptor. Then, the extracted features are clustered by the well-known K-means algorithm. The experimental results demonstrate the effectiveness of the proposed method compared to state-of-the-art approaches in several public benchmark datasets. The application of the proposed method and additional experimental results for several challenging datasets are available online.},
}
@article{vandenEnde:2019a,
abbr = {},
bibtex_show = {true},
author = {van den Ende, R.P.J. and Kerkhof, E.M. and Rigter, L.S. and van Leerdam, M.E. and Peters, F.P. and van Triest, B. and Staring, M. and Marijnen, C.A.M. and van der Heide, U.A.},
title = {Feasibility of gold fiducial markers as a surrogate for GTV position in image-guided radiotherapy of rectal cancer},
journal = {International Journal of Radiation Oncology, Biology, Physics},
volume = {105},
number = {5},
pages = {1151 -- 1159},
month = {December},
year = {2019},
pdf = {2019_j_IJROBP.pdf},
html = {https://doi.org/10.1016/j.ijrobp.2019.08.052},
arxiv = {},
code = {},
abstract = {<b>Purpose:</b> To evaluate the feasibility of fiducial markers as a surrogate for GTV position in image-guided radiotherapy of rectal cancer.<br><b>Methods and Materials:</b> We analyzed 35 fiducials in 19 rectal cancer patients who received short course radiotherapy or long-course chemoradiotherapy. A MRI exam was acquired before and after the first week of radiotherapy and daily pre- and post-irradiation CBCT scans were acquired in the first week of radiotherapy. Between the two MRI exams, the fiducial displacement relative to the center of gravity of the GTV (COGGTV) and the COGGTV displacement relative to bony anatomy was determined. Using the CBCT scans, inter- and intrafraction fiducial displacement relative to bony anatomy was determined.<br><b>Results:</b> The systematic error of the fiducial displacement relative to the COG<sub>GTV</sub> was 2.8, 2.4 and 4.2 mm in the left-right (LR), anterior-posterior (AP) and craniocaudal (CC) direction. Large interfraction systematic errors of up to 8.0 and random errors up to 4.7 mm were found for COG<sub>GTV</sub> and fiducial displacements relative to bony anatomy, mostly in the AP and CC directions. For tumors located in the mid- and upper rectum these errors were up to 9.4 (systematic) and 5.6 mm (random) compared to 4.9 and 2.9 mm for tumors in the lower rectum. Systematic and random errors of the intrafraction fiducial displacement relative to bony anatomy were ≤ 2.1 mm in all directions.<br><b>Conclusions:</b> Large interfraction errors of the COG<sub>GTV</sub> and the fiducials relative to bony anatomy were found. Therefore, despite the observed fiducial displacement relative to the COG<sub>GTV</sub>, the use of fiducials as a surrogate for GTV position reduces the required margins in the AP and CC direction for a GTV boost using image-guided radiotherapy of rectal cancer. This reduction may be larger in patients with tumors located in the mid- and upper rectum compared to the lower rectum.},
}
@article{Qiao:2019a,
abbr = {},
bibtex_show = {true},
author = {Qiao, Yuchuan and Jagt, Thyrza and Hoogeman, Mischa and Lelieveldt, Boudewijn P.F. and Staring, Marius},
title = {Evaluation of an open source registration package for automatic contour propagation in online adaptive intensity-modulated proton therapy of prostate cancer},
journal = {Frontiers in Oncology},
volume = {9},
pages = {1297},
month = {November},
year = {2019},
pdf = {2019_j_FiO.pdf},
html = {https://doi.org/10.3389/fonc.2019.01297},
arxiv = {},
code = {https://github.com/SuperElastix/elastix},
abstract = {<b>Objective:</b> Our goal was to investigate the performance of an open source deformable image registration package, elastix, for fast and robust contour propagation in the context of online adaptive intensity-modulated proton therapy (IMPT) for prostate cancer.<br><b>Methods:</b> A planning and 7-10 repeat CT scans were available of 18 prostate cancer patients. Automatic contour propagation of repeat CT scans was performed using elastix and compared with manual delineations in terms of geometric accuracy and runtime. Dosimetric accuracy was quantified by generating IMPT plans using the propagated contours expanded with a 2 mm (prostate) and 3.5 mm margin (seminal vesicles and lymph nodes) and calculating dosimetric coverage based on the manual delineation. A coverage of V95% ≥ 98% (at least 98% of the target volumes receive at least 95% of the prescribed dose) was considered clinically acceptable.<br><b>Results:</b> Contour propagation runtime varied between 3 and 30 seconds for different registration settings. For the fastest setting, 83 in 93 (89.2%), 73 in 93 (78.5%), and 91 in 93 (97.9%) registrations yielded clinically acceptable dosimetric coverage of the prostate, seminal vesicles, and lymph nodes, respectively. For the prostate, seminal vesicles, and lymph nodes the Dice Similarity Coefficient (DSC) was 0:87 ± 0:05, 0:63 ± 0:18 and 0:89 ± 0:03 and the mean surface distance (MSD) was 1:4 ± 0:5 mm, 2:0 ± 1:2 mm and 1:5 ± 0:4 mm, respectively.<br><b>Conclusion:</b> With a dosimetric success rate of 78.5% to 97.9%, this software may facilitate online adaptive IMPT of prostate cancer using a fast, free and open implementation.},
}
@article{Zhai:2019a,
abbr = {},
bibtex_show = {true},
author = {Zhai, Zhiwei and Staring, Marius and Ninaber, Maarten K. and de Vries-Bouwstra, Jeska and Schouffoer, Anne A. and Kroft, Lucia J. and Stolk, Jan and Stoel, Berend C.},
title = {Pulmonary Vascular Morphology Associated with Gas Exchange in Systemic Sclerosis without Lung Fibrosis},
journal = {Journal of Thoracic Imaging},
volume = {34},
number = {6},
pages = {373 -- 379},
month = {November},
year = {2019},
pdf = {2019_j_JTI.pdf},
html = {http://dx.doi.org/10.1097/RTI.0000000000000395},
arxiv = {},
code = {},
abstract = {<b>Purpose:</b> Gas exchange in systemic sclerosis (SSc) is known to be affected by fibrotic changes in the pulmonary parenchyma. However, SSc patients without detectable fibrosis can still have impaired gas transfer. We aim to investigate whether pulmonary vascular changes could partly explain a reduction in gas transfer of systemic sclerosis (SSc) patients without fibrosis.<br><b>Materials and Methods:</b> We selected 77 patients, whose visual CT scoring showed no fibrosis. Pulmonary vessels were detected automatically in CT images and their local radii were calculated. The frequency of occurrence for each radius was calculated, and from this radius histogram two imaging biomarkers (α and β) were extracted, where α reflects the relative contribution of small vessels compared to large vessels and β represents the vessel tree capacity. Correlations between imaging biomarkers and gas transfer (DLCOc %predicted) were evaluated with Spearman's correlation. Multivariable stepwise linear regression was performed with DLCOc %predicted as dependent variable and age, BMI, sPAP, FEV1 %predicted, TLC %predicted, FVC %predicted, α, β, voxel size and CT-derived lung volume as independent variables.<br><b>Results:</b> Both α and β were significantly correlated with gas transfer (R=-0.29, p-value=0.011 and R=0.32, p-value=0.004, respectively). The multivariable step-wise linear regression analysis selected sPAP (coefficient=-0.78, 95%CI=[-1.07, -0.49], p-value<0.001), β (coefficient=8.6, 95%CI=[4.07, 13.1], p-value<0.001) and FEV1 %predicted (coefficient=0.3, 95%CI=[0.12, 0.48], p-value=0.001) as significant independent predictors of DLCOc %predicted (R=0.71, p-value<0.001).<br><b>Conclusions:</b>In SSc patients without detectable pulmonary fibrosis, pulmonary vascular morphology is associated with gas transfer, indicating that impaired gas exchange is associated with vascular changes.},
}
@article{Qiao:2019b,
abbr = {TMI},
bibtex_show = {true},
author = {Qiao, Yuchuan and Lelieveldt, Boudewijn P.F and Staring, Marius},
title = {An efficient preconditioner for stochastic gradient descent optimization of image registration},
journal = {IEEE Transactions on Medical Imaging},
volume = {38},
number = {10},
pages = {2314 -- 2325},
month = {October},
year = {2019},
pdf = {2019_j_TMI.pdf},
html = {https://doi.org/10.1109/TMI.2019.2897943},
arxiv = {},
code = {https://github.com/SuperElastix/elastix},
abstract = {Stochastic gradient descent (SGD) is commonly used to solve (parametric) image registration problems. In case of badly scaled problems, SGD however only exhibits sublinear convergence properties. In this paper we propose an efficient preconditioner estimation method to improve the convergence rate of SGD. Based on the observed distribution of voxel displacements in the registration, we estimate the diagonal entries of a preconditioning matrix, thus rescaling the optimization cost function. The preconditioner is efficient to compute and employ, and can be used for mono-modal as well as multi-modal cost functions, in combination with different transformation models like the rigid, affine and B-spline model. Experiments on different clinical data sets show that the proposed method indeed improves the convergence rate compared to SGD with speedups around 2-5 in all tested settings, while retaining the same level of registration accuracy.},
}
@article{Bayer:2019,
abbr = {},
bibtex_show = {true},
author = {Bayer, Siming and Zhai, Zhiwei and Strumia, Maddalena and Tong, Xiaoguang and Gao, Ying and Staring, Marius and Stoel, Berend and Fahrig, Rebecca and Nabavi, Arya and Maier, Andreas and Ravikumar, Nishant},
title = {Registration of vascular structures using a hybrid mixture model},
journal = {International Journal of Computer Assisted Radiology and Surgery},
volume = {14},
number = {9},
pages = {1507 -- 1516},
month = {September},
year = {2019},
pdf = {2019_j_IJCARS.pdf},
html = {https://doi.org/10.1007/s11548-019-02007-y},
arxiv = {},
code = {},
abstract = {<b>Purpose:</b> Morphological changes to anatomy resulting from invasive surgical procedures or pathology, typically alter the surrounding vasculature. This makes it useful as a descriptor for feature-driven image registration in various clinical applications. However, registration of vasculature remains challenging, as vessels often differ in size and shape, and may even miss branches, due to surgical interventions or pathological changes. Furthermore, existing vessel registration methods are typically designed for a specific application. To address this limitation, we propose a generic vessel registration approach useful for a variety of clinical applications, involving different anatomical regions.<br><b>Methods:</b> A probabilistic registration framework based on a hybrid mixture model, with a refinement mechanism to identify missing branches (denoted as HdMM+) during vasculature matching, is introduced. Vascular structures are represented as 6-dimensional hybrid point sets comprising spatial positions and centerline orientations, using Student's t-distributions to model the former and Watson distributions for the latter.<br><b>Results:</b> The proposed framework is evaluated for intraoperative brain shift compensation, and monitoring changes in pulmonary vasculature resulting from chronic lung disease. Registration accuracy is validated using both synthetic and patient data. Our results demonstrate, HdMM+ is able to reduce more than 85% of the initial error for both applications, and outperforms the state-of-the-art point-based registration methods such as coherent point drift (CPD) and Student's t-Distribution mixture model (TMM), in terms of mean surface distance, modified hausdorff distance, Dice and Jaccard scores.<br><b>Conclusion:</b> The proposed registration framework models complex vascular structures using a hybrid representation of vessel centerlines, and accommodates intricate variations in vascular morphology. Furthermore, it is generic and flexible in its design, enabling its use in a variety of clinical applications.},
}
@article{Zhai:2019b,
abbr = {MP},
bibtex_show = {true},
author = {Zhai, Zhiwei and Staring, Marius and Giron, Irene Hernandez and Veldkamp, Wouter J.H. and Kroft, Lucia J. and Ninaber, Maarten K. and Stoel, Berend C.},
title = {Automatic quantitative analysis of pulmonary vascular morphology in CT images},
journal = {Medical Physics},
volume = {46},
number = {9},
pages = {3985 -- 3997},
month = {September},
year = {2019},
pdf = {2019_j_MPb.pdf},
html = {https://doi.org/10.1002/mp.13659},
arxiv = {},
code = {},
abstract = {<b>Purpose:</b> Vascular remodeling is a significant pathological feature of various pulmonary diseases, which may be assessed by quantitative CT imaging. The purpose of this study was therefore to develop and validate an automatic method for quantifying pulmonary vascular morphology in CT images.<br><b>Methods:</b> The proposed method consists of pulmonary vessel extraction and quantification. For extracting pulmonary vessels, a graph-cuts based method is proposed which considers appearance (CT intensity) and shape (vesselness from a Hessian-based filter) features, and incorporates distance to the airways into the cost function to prevent false detection of airway walls. For quantifying the extracted pulmonary vessels, a radius histogram is generated by counting the occurrence of vessel radii, calculated from a distance transform based method. Subsequently, two biomarkers, slope α and intercept β, are calculated by linear regression on the radius histogram. A public data set from the VESSEL12 challenge was used to independently evaluate the vessel extraction. The quantitative analysis method was validated using images of a 3D printed vessel phantom, scanned by a clinical CT scanner and a micro-CT scanner (to obtain a gold standard). To confirm the association between imaging biomarkers and pulmonary function, 77 scleroderma patients were investigated with the proposed method.<br><b>Results:</b> In the independent evaluation with the public data set, our vessel segmentation method obtained an area under the ROC curve of 0.976. The median radius difference between clinical and micro-CT scans of a 3D printed vessel phantom was 0.062 ± 0.020 mm, with interquartile range of 0.199 ± 0.050 mm. In the studied patient group, a significant correlation between diffusion capacity for carbon monoxide and the biomarkers, α (R=-0.27, p-value=0.018) and β (R=0.321, p-value=0.004), was obtained.<br><b>Conclusions:</b> In conclusion, the proposed method was highly accurate, validated with a public data set and a 3D printed vessel phantom data set. The correlation between imaging biomarkers and diffusion capacity in a clinical data set confirmed an association between lung structure and function. This quantification of pulmonary vascular morphology may be helpful in understanding the pathophysiology of pulmonary vascular diseases.},
}
@article{Elmahdy:2019,
abbr = {MP},
bibtex_show = {true},
author = {Elmahdy, Mohamed S. and Jagt, Thyrza and Zinkstok, Roel Th. and Qiao, Yuchuan and Shazad, Rahil and Sokooti, Hessam and Yousefi, Sahar and Incrocci, Luca and Marijnen, Corrie A.M. and Hoogeman, Mischa and Staring, Marius},
title = {Robust contour propagation using deep learning and image registration for online adaptive proton therapy of prostate cancer},
journal = {Medical Physics},
volume = {46},
number = {8},
pages = {3329 -- 3343},
month = {August},
year = {2019},
pdf = {2019_j_MPa.pdf},
html = {https://doi.org/10.1002/mp.13620},
arxiv = {},
code = {},
abstract = {<b>Purpose:</b> To develop and validate a robust and accurate registration pipeline for automatic contour propagation for online adaptive Intensity-Modulated Proton Therapy (IMPT) of prostate cancer using elastix software and deep learning.<br><b>Methods:</b> A 3D Convolutional Neural Network was trained for automatic bladder segmentation of the CT scans. The automatic bladder segmentation alongside the CT scan are jointly optimized to add explicit knowledge about the underlying anatomy to the registration algorithm. We included three datasets from different institutes and CT manufacturers. The first was used for training and testing the ConvNet, where the second and the third were used for evaluation of the proposed pipeline. The system performance was quantified geometrically using the Dice Similarity Coefficient (DSC), the Mean Surface Distance (MSD), and the 95% Hausdorff Distance (HD). The propagated contours were validated clinically through generating the associated IMPT plans and compare it with the IMPT plans based on the manual delineations. Propagated contours were considered clinically acceptable if their treatment plans met the dosimetric coverage constraints on the manual contours.<br><b>Results:</b> The bladder segmentation network achieved a DSC of 88% and 82% on the test datasets. The proposed registration pipeline achieved a MSD of 1.29 ± 0.39, 1.48 ± 1.16, and 1.49 ± 0.44 mm for the prostate, seminal vesicles, and lymph nodes, respectively on the second dataset and a MSD of 2.31 ± 1.92 and 1.76 ± 1.39 mm for the prostate and seminal vesicles on the third dataset. The automatically propagated contours met the dose coverage constraints in 86%, 91%, and 99% of the cases for the prostate, seminal vesicles, and lymph nodes, respectively. A Conservative Success Rate (CSR) of 80% was obtained, compared to 65% when only using intensity-based registration.<br><b>Conclusions:</b> The proposed registration pipeline obtained highly promising results for generating treatment plans adapted to the daily anatomy. With 80% of the automatically generated treatment plans directly usable without manual correction, a substantial improvement in system robustness was reached compared to a previous approach. The proposed method therefore facilitates more precise proton therapy of prostate cancer, potentially leading to fewer treatment related adverse side effects.},
}
@article{Sokooti:2019,
abbr = {},
bibtex_show = {true},
author = {Sokooti, Hessam and Saygili, Gorkem and Glocker, Ben and Lelieveldt, Boudewijn P.F. and Staring, Marius},
title = {Quantitative Error Prediction of Medical Image Registration using Regression Forests},
journal = {Medical Image Analysis},
volume = {56},
number = {8},
pages = {110 -- 121},
month = {August},
year = {2019},
pdf = {2019_j_MedIAb.pdf},
html = {https://doi.org/10.1016/j.media.2019.05.005},
arxiv = {},
code = {},
abstract = {Predicting registration error can be useful for evaluation of registration procedures, which is important for the adoption of registration techniques in the clinic. In addition, quantitative error prediction can be helpful in improving the registration quality. The task of predicting registration error is demanding due to the lack of a ground truth in medical images. This paper proposes a new automatic method to predict the registration error in a quantitative manner, and is applied to chest CT scans. A random regression forest is utilized to predict the registration error locally. The forest is built with features related to the transformation model and features related to the dissimilarity after registration. The forest is trained and tested using manually annotated corresponding points between pairs of chest CT scans in two experiments: SPREAD (trained and tested on SPREAD) and inter-database (including three databases SPREAD, DIR-Lab-4DCT and DIR-Lab-COPDgene). The results show that the mean absolute errors of regression are 1.07 ± 1.86 and 1.76 ± 2.59 mm for the SPREAD and inter-database experiment, respectively. The overall accuracy of classification in three classes (correct, poor and wrong registration) is 90.7% and 75.4%, for SPREAD and inter-database respectively. The good performance of the proposed method enables important applications such as automatic quality control in large-scale image analysis.},
}
@article{vandenEnde:2019b,
abbr = {},
bibtex_show = {true},
author = {van den Ende, R.P.J. and Rigter, L.S. and Kerkhof, E.M. and van Persijn van Meerten, E.L. and Rijkmans, E.C. and Lambregts, D.M.J. and van Triest, B. and van Leerdam, M.E. and Staring, M. and Marijnen, C.A.M. and van der Heide, U.A.},
title = {MRI visibility of gold fiducial markers for image-guided radiotherapy of rectal cancer},
journal = {Radiotherapy & Oncology},
volume = {132},
number = {3},
pages = {93 -- 99},
month = {March},
year = {2019},
pdf = {2019_j_RO.pdf},
html = {https://doi.org/10.1016/j.radonc.2018.11.016},
arxiv = {},
code = {},
abstract = {<b>Background and purpose:</b> A GTV boost is suggested to result in higher complete response rates in rectal cancer patients, which is attractive for organ preservation. Fiducials may offer GTV position verification on (CB)CT, if the fiducial-GTV spatial relationship can be accurately defined on MRI. The study aim was to evaluate the MRI visibility of fiducials inserted in the rectum.<br><b>Materials and methods:</b> We tested four fiducial types (two Visicoil types, Cook and Gold Anchor), inserted in five patients each. Four observers identified fiducial locations on two MRI exams per patient in two scenarios: without (scenario A) and with (scenario B) (CB)CT available. A fiducial was defined to be consistently identified if 3 out of 4 observers labeled that fiducial at the same position on MRI. Fiducial visibility was scored on an axial and sagittal T2-TSE sequence and a T1 3D GRE sequence.<br><b>Results:</b> Fiducial identification was poor in scenario A for all fiducial types. The Visicoil 0.75 and Gold Anchor were the most consistently identified fiducials in scenario B with 7 out of 9 and 8 out of 11 consistently identified fiducials in the first MRI exam and 2 out of 7 and 5 out of 10 in the second MRI exam, respectively. The consistently identified Visicoil 0.75 and Gold Anchor fiducials were best visible on the T1 3D GRE sequence.<br><b>Conclusion:</b> The Visicoil 0.75 and Gold Anchor fiducials were the most visible fiducials on MRI as they were most consistently identified. The use of a registered (CB)CT and a T1 3D GRE MRI sequence is recommended.},
}
@article{DeVos:2019,
abbr = {},
bibtex_show = {true},
author = {De Vos, Bob and Berendsen, Floris F. and Viergever, Max A. and Sokooti, Hessam and Staring, Marius and I{\v{s}}gum, Ivana},
title = {A Deep Learning Framework for Unsupervised Affine and Deformable Image Registration},
journal = {Medical Image Analysis},
volume = {52},
number = {2},
pages = {128 -- 143},
month = {February},
year = {2019},
pdf = {2019_j_MedIAa.pdf},
html = {https://doi.org/10.1016/j.media.2018.11.010},
arxiv = {},
code = {},
abstract = {Image registration, the process of aligning two or more images, is the core technique of many (semi-)automatic medical image analysis tasks. Recent studies have shown that deep learning methods, notably convolutional neural networks (ConvNets), can be used for image registration. Thus far training of ConvNets for registration was supervised using predefined example registrations. However, obtaining example registrations is not trivial. To circumvent the need for predefined examples, and thereby to increase convenience of training ConvNets for image registration, we propose the Deep Learning Image Registration (DLIR) framework for unsupervised affine and deformable image registration. In the DLIR framework ConvNets are trained for image registration by exploiting image similarity analogous to conventional intensity-based image registration. After a ConvNet has been trained with the DLIR framework, it can be used to register pairs of unseen images in one shot. We propose flexible ConvNets designs for affine image registration and for deformable image registration. By stacking multiple of these ConvNets into a larger architecture, we are able to perform coarse-to-fine image registration. We show for registration of cardiac cine MRI and registration of chest CT that performance of the DLIR framework is comparable to conventional image registration while being several orders of magnitude faster.},
}
@article{Sun:2018,
abbr = {},
bibtex_show = {true},
author = {Sun, Zhuo and Qiao, Yuchuan and Lelieveldt, Boudewijn P.F. and Staring, Marius},
title = {Integrating Spatial-Anatomical Regularization and Structure Sparsity into SVM: Improving Interpretation of Alzheimer's Disease Classification},
journal = {NeuroImage},
volume = {178},
pages = {445 -- 460},
month = {September},
year = {2018},
pdf = {2018_j_NI.pdf},
html = {https://doi.org/10.1016/j.neuroimage.2018.05.051},
arxiv = {},
code = {},
abstract = {In recent years, machine learning approaches have been successfully applied to the field of neuroimaging for classification and regression tasks. However, many approaches do not give an intuitive relation between the raw features and the diagnosis. Therefore, they are difficult for clinicians to interpret. Moreover, most approaches treat the features extracted from the brain (for example, voxelwise gray matter concentration maps from brain MRI) as independent variables and ignore their spatial and anatomical relations. In this paper, we present a new Support Vector Machine (SVM)-based learning method for the classification of Alzheimer's disease (AD), which integrates spatial-anatomical information. In this way, spatial-neighbor features in the same anatomical region are encouraged to have similar weights in the SVM model. Secondly, to make the learned model more interpretable, we introduce a group lasso penalty to induce structure sparsity, which may help clinicians to assess the key regions involved in the disease. For solving this learning problem, we use an accelerated proximal gradient descent approach. We tested our method on the subset of ADNI data selected by Cuingnet et al. (2011) for Alzheimer's disease classification, as well as on an independent larger dataset from ADNI. Good classification performance is obtained for distinguishing cognitive normals (CN) vs. AD, as well as on distinguishing between various sub-types (e.g. CN vs. Mild Cognitive Impairment). The model trained on Cuignet's dataset for AD vs. CN classification was directly used without re-training to the independent larger dataset. Good performance was achieved, demonstrating the generalizability of the proposed methods. For all experiments, the classification results are comparable or better than the state-of-the-art, while the weight map more clearly indicates the key regions related to Alzheimer's disease.},
}
@article{Zhai:2018,
abbr = {},
bibtex_show = {true},
author = {Zhai, Zhiwei and Ota, Hideki and Staring, Marius and Stolk, Jan and Sugimura, Koichiro and Takase, Kei and Stoel, Berend C.},
title = {Treatment Effect of Balloon Pulmonary Angioplasty in Chronic Thromboembolic Pulmonary Hypertension Quantified by Automatic Comparative Imaging in Computed Tomography Pulmonary Angiography},
journal = {Investigative Radiology},
volume = {53},
number = {5},
pages = {286 -- 292},
month = {May},
year = {2018},
pdf = {2018_j_IR.pdf},
html = {https://doi.org/10.1097/RLI.0000000000000441},
arxiv = {},
code = {},
abstract = {<b>Objectives:</b> Balloon pulmonary angioplasty (BPA) in patients with inoperable chronic thromboembolic pulmonary hypertension (CTEPH) can have variable outcomes. To gain more insight into this variation, we designed a method for visualizing and quantifying changes in pulmonary perfusion by automatically comparing computed tomography (CT) pulmonary angiography before and after BPA treatment. We validated these quantifications of perfusion changes against hemodynamic changes measured with right-sided heart catheterization.<br><b>Materials and Methods:</b> We studied 14 consecutive CTEPH patients (12 women; age, 70.5 ± 24), who underwent CT pulmonary angiography and right-sided heart catheterization, before and after BPA. Posttreatment images were registered to pretreatment CT scans (using the Elastix toolbox) to obtain corresponding locations. Pulmonary vascular trees and their centerlines were detected using a graph cuts method and a distance transform method, respectively. Areas distal from vessels were defined as pulmonary parenchyma. Subsequently, the density changes within the vascular centerlines and parenchymal areas were calculated and corrected for inspiration level differences. For visualization, the densitometric changes were displayed in color-coded overlays. For quantification, the median and interquartile range of the density changes in the vascular and parenchymal areas (ΔVD and ΔPD) were calculated. The recorded changes in hemodynamic parameters, including changes in systolic, diastolic, mean pulmonary artery pressure (ΔsPAP, ΔdPAP and ΔmPAP, respectively) and vascular resistance (ΔPVR), were used as reference assessments of the treatment effect. Spearman correlation coefficients were employed to investigate the correlations between changes in perfusion and hemodynamic changes.<br><b>Results:</b> Comparative imaging maps showed distinct patterns in perfusion changes among patients. Within pulmonary vessels, the interquartile range of ΔVD correlated significantly with ΔsPAP (R= 0.58, p=0.03), ΔdPAP (R= 0.71, p=0.005), ΔmPAP (R= 0.71, p=0.005), and ΔPVR (R= 0.77, p=0.001). In the parenchyma, the median of ΔPD had significant correlations with ΔdPAP (R= 0.58, p=0.030) and ΔmPAP (R= 0.59, p=0.025).<br><b>Conclusions:</b> Comparative imaging analysis in CTEPH patients offers insight into differences in BPA treatment effect. Quantification of perfusion changes provides noninvasive measures that reflect hemodynamic changes.},
}
@article{Shahzad:2017,
abbr = {},
bibtex_show = {true},
author = {Shahzad, Rahil and Tao, Qian and Dzyubachyk, Oleh and Staring, Marius and Lelieveldt, Boudewijn P.F. and van der Geest, Rob J.},
title = {Fully-Automatic Left Ventricular Segmentation from Long-Axis Cardiac Cine MR Scans},
journal = {Medical Image Analysis},
volume = {39},
pages = {44 -- 55},
month = {July},
year = {2017},
pdf = {2017_j_MedIA.pdf},
html = {http://doi.org/10.1016/j.media.2017.04.004},
arxiv = {},
code = {},
abstract = {With an increasing number of large-scale population-based cardiac magnetic resonance (CMR) imaging studies being conducted nowadays, there comes the mammoth task of image annotation and image analysis. Such population-based studies would greatly benefit from automated pipelines, with an efficient CMR image analysis workflow. The purpose of this work is to investigate the feasibility of using a fully-automatic pipeline to segment the left ventricular endocardium and epicardium simultaneously on two orthogonal (vertical and horizontal) long-axis cardiac cine MRI scans. The pipeline is based on a multi-atlas-based segmentation approach and a spatio-temporal registration approach. The performance of the method was assessed by: (i) comparing the automatic segmentations to those obtained manually at both the end-diastolic and end-systolic phase, (ii) comparing the automatically obtained clinical parameters, including end-diastolic volume, end-systolic volume, stroke volume and ejection fraction, with those defined manually and (iii) by the accuracy of classifying subjects to the appropriate risk category based on the estimated ejection fraction. Automatic segmentation of the left ventricular endocardium was achieved with a Dice similarity coefficient (DSC) of 0.93 on the end-diastolic phase for both the vertical and horizontal long-axis scan; on the end-systolic phase the DSC was 0.88 and 0.85, respectively. For the epicardium, a DSC of 0.94 and 0.95 was obtained on the end-diastolic vertical and horizontal long-axis scans; on the end-systolic phase the DSC was 0.90 and 0.88, respectively. With respect to the clinical volumetric parameters, Pearson correlation coefficient (R) of 0.97 was obtained for the end-diastolic volume, 0.95 for end-systolic volume, 0.87 for stroke volume and 0.84 for ejection fraction. Risk category classification based on ejection fraction showed that 80% of the subjects were assigned to the correct risk category and only one subject (< 1%) was more than one risk category off. We conclude that the proposed automatic pipeline presents a viable and cost-effective alternative for manual annotation.},
}
@article{Sun:2017,
abbr = {},
bibtex_show = {true},
author = {Sun, Zhuo and van der Giessen, Martijn and Lelieveldt, Boudewijn P.F. and Staring, Marius},
title = {Detection of conversion from mild cognitive impairment to Alzheimer's disease using longitudinal brain MRI},
journal = {Frontiers in Neuroinformatics},
volume = {11},
pages = {16},
month = {February},
year = {2017},
pdf = {2017_j_FNI.pdf},
html = {http://dx.doi.org/10.3389/fninf.2017.00016},
arxiv = {},
code = {},
abstract = {Mild Cognitive Impairment (MCI) is an intermediate stage between healthy and Alzheimer's disease (AD). To enable early intervention it is important to identify the MCI subjects that will convert to AD in an early stage. In this paper, we provide a new method to distinguish between MCI patients that either convert to Alzheimer's Disease (MCIc) or remain stable (MCIs), using only longitudinal T1-weighted MRI. Currently, most longitudinal studies focus on volumetric comparison of a few anatomical structures, thereby ignoring more detailed development inside and outside those structures. In this study we propose to exploit the anatomical development within the entire brain, as found by a non-rigid registration approach. Specifically, this anatomical development is represented by the stationary velocity field (SVF) from registration between the baseline and follow-up images. To make the SVFs comparable among subjects, we use the parallel transport method to align them in a common space. The normalized SVF together with derived features are then used to distinguish between MCIc and MCIs subjects. This novel feature space is reduced using a Kernel Principal Component Analysis method, and a linear support vector machine is used as a classifier. Extensive comparative experiments are performed to inspect the influence of several aspects of our method on classification performance, specifically the feature choice, the smoothing parameter in the registration and the use of dimensionality reduction. The optimal result from a 10-fold cross-validation using 36 month follow-up data shows competitive results: accuracy 92%, sensitivity 95%, specificity 90%, and AUC 94%. Based on the same dataset, the proposed approach outperforms two alternative ones that either depends on the baseline image only, or uses longitudinal information from larger brain areas. Good results were also obtained when scans at 6, 12 or 24 months were used for training the classifier. Besides the classification power, the proposed method can quantitatively compare brain regions that have a significant difference in development between the MCIc and MCIs groups.},
}
@article{Dzyubachyk:2017,
abbr = {},
bibtex_show = {true},
author = {Dzyubachyk, Oleh and Staring, Marius and Reijnierse, Monique and Lelieveldt, Boudewijn P.F. and van der Geest, Rob J.},
title = {Inter-Station Intensity Standardization for Whole-Body MR Data},
journal = {Magnetic Resonance in Medicine},
volume = {77},
number = {1},
pages = {422 -- 433},
month = {January},
year = {2017},
pdf = {2017_j_MRM.pdf},
html = {http://dx.doi.org/10.1002/mrm.26098},
arxiv = {},
code = {},
abstract = {<b>Purpose.</b> To develop and validate a method for performing inter-station intensity standardization in multi-spectral whole-body MR data.<br><b>Methods.</b> Different approaches for mapping the intensity of each acquired image stack into the reference intensity space were developed and validated. The registration strategies included: "direct" registration to the reference station (Strategy 1), "progressive" registration to the neighbouring stations without (Strategy 2) and with (Strategy 3) using information from the overlap regions of the neighbouring stations. For Strategy 3, two regularized modifications were proposed and validated. All methods were tested on two multi-spectral whole-body MR data sets: a multiple myeloma patients data set (48 subjects) and a whole-body MR angiography data set (33 subjects).<br><b>Results.</b> For both data sets, all strategies showed significant improvement of intensity homogeneity with respect to vast majority of the validation measures (p < 0.005). Strategy 1 exhibited the best performance, closely followed by Strategy 2. Strategy 3 and its modifications were performing worse, in majority of the cases significantly (p < 0.05).<br><b>Conclusions.</b> We propose several strategies for performing inter-station intensity standardization in multi-spectral whole-body MR data. All the strategies were successfully applied to two types of whole-body MR data, and the "direct" registration strategy was concluded to perform the best.},
}
@article{Viergever:2016,
abbr = {},
bibtex_show = {true},
author = {Viergever, Max A. and Maintz, J.B. Antoine and Klein, Stefan and Murphy, Keelin and Staring, Marius and Pluim, Josien P.W.},
title = {A survey of medical image registration - under review},
journal = {Medical Image Analysis},
volume = {33},
pages = {140 -- 144},
month = {October},
year = {2016},
pdf = {2016_j_MedIA.pdf},
html = {http://dx.doi.org/10.1016/j.media.2016.06.030},
arxiv = {},
code = {},
abstract = {A retrospective view on the past two decades of the field of medical image registration is presented, guided by the article "A survey of medical image registration" (Maintz and Viergever, 1998). It shows that the classification of the field introduced in that article is still usable, although some modifications to do justice to advances in the field would be due. The main changes over the last twenty years are the shift from extrinsic to intrinsic registration, the primacy of intensity-based registration, the breakthrough of nonlinear registration, the progress of inter-subject registration, and the availability of generic image registration software packages. Two problems that were called urgent already 20 years ago, are even more urgent nowadays: Validation of registration methods, and translation of results of image registration research to clinical practice. It may be concluded that the field of medical image registration has evolved, but still is in need of further development in various aspects.},
}
@article{Xiao:2016,
abbr = {},
bibtex_show = {true},
author = {Xiao, Changyan and Stoel, Berend C. and Bakker, M. Els and Peng, Yuanyuan and Stolk, Jan and Staring, Marius},
title = {Pulmonary Fissure Detection in CT Images Using a Derivative of Stick Filter},
journal = {IEEE Transactions on Medical Imaging},
volume = {35},
number = {6},
pages = {1488 -- 1500},
month = {June},
year = {2016},
pdf = {2016_j_TMIc.pdf},
html = {http://dx.doi.org/10.1109/TMI.2016.2517680},
arxiv = {},
code = {},
abstract = {Pulmonary fissures are important landmarks for recognition of lung anatomy. In CT images, automatic detection of fissures is complicated by factors like intensity variability, pathological deformation and imaging noise. To circumvent this problem, we propose a derivative of stick (DoS) filter for fissure enhancement and a post-processing pipeline for subsequent segmentation. Considering a typical thin curvilinear shape of fissure profiles inside 2D cross-sections, the DoS filter is presented by first defining nonlinear derivatives along a triple stick kernel in varying directions. Then, to accommodate pathological abnormality and orientational deviation, a max-min cascading and multiple plane integration scheme is adopted to form a shape-tuned likelihood for 3D surface patches discrimination. During the post-processing stage, our main contribution is to isolate the fissure patches from adhering clutters by introducing a branch-point removal algorithm, and a multi-threshold merging framework is employed to compensate for local intensity inhomogeneity. The performance of our method was validated in experiments with two clinical CT data sets including 55 publicly available LOLA11 scans as well as separate left and right lung images from 23 GLUCOLD scans of COPD patients. Compared with manually delineating interlobar boundary references, our method obtained a high segmentation accuracy with median F1-scores of 0.833, 0.885, and 0.856 for the LOLA11, left and right lung images respectively, whereas the corresponding indices for a conventional Wiemker filtering method were 0.687, 0.853, and 0.841. The good performance of our proposed method was also verified by visual inspection and demonstration on abnormal and pathological cases, where typical deformations were robustly detected together with normal fissures.},
}
@article{Saygili:2016,
abbr = {},
bibtex_show = {true},
author = {Saygili, Gorkem and Staring, Marius and Hendriks, Emile A.},
title = {Confidence Estimation for Medical Image Registration Based On Stereo Confidences},
journal = {IEEE Transactions on Medical Imaging},
volume = {35},
number = {2},
pages = {539 -- 549},
month = {February},
year = {2016},
pdf = {2016_j_TMIb.pdf},
html = {http://dx.doi.org/10.1109/TMI.2015.2481609},
arxiv = {},
code = {},
abstract = {In this paper, we propose a novel method to estimate the confidence of a registration that does not require any ground truth, is independent from the registration algorithm and the resulting confidence is correlated with the amount of registration error. We first apply a local search to match patterns between the registered image pairs. Local search induces a cost space per voxel which we explore further to estimate the confidence of the registration similar to confidence estimation algorithms for stereo matching. We test our method on both synthetically generated registration errors and on real registrations with ground truth. The experimental results show that our confidence measure can estimate registration errors and it is correlated with local errors.},
}
@article{Qiao:2016,
abbr = {},
bibtex_show = {true},
author = {Qiao, Yuchuan and van Lew, Baldur and Lelieveldt, Boudewijn P.F and Staring, Marius},
title = {Fast Automatic Step Size Estimation for Gradient Descent Optimization of Image Registration},
journal = {IEEE Transactions on Medical Imaging},
volume = {35},
number = {2},
pages = {391 -- 403},
month = {February},
year = {2016},
pdf = {2016_j_TMIa.pdf},
html = {http://dx.doi.org/10.1109/TMI.2015.2476354},