This repository accompanies the journal submission of "Benchmarking of Deep Learning Methods for Generic MRI Multi-Organ Abdominal Segmentation", with authors Deepa Krishnaswamy*, Cosmin Ciausu*, Steve Pieper, Ron Kikinis, Benjamin Billot, Andrey Fedorov. (*shared first author)
In this repository, we:
- Provide the model weights for ABDSynth, which is a model based on SynthSeg (https://github.com/BBillot/SynthSeg) trained using abdominal data, and
- Provide a demo notebook to run inference on a sample dataset using ABDSynth
- Provide our preprocessing scripts for AMOS, CHAOS, and LiverHCCSeg
- Provide our preprocessed data used for evaluation as a GitHub release attachment.
Further instructions for the demo notebook:
- Please install SynthSeg (https://github.com/BBillot/SynthSeg) for python 3.8
- Additionally, install Simple ITK
In this publication, we:
- Perform a benchmarking study of four state-of-the-art algorithms for segmentation of abdominal structures in MR images, including three pre-trained models, and another that we train, termed ABDSynth
- Perform an analysis on three different datasets containing both normal and pathological cases of a variety of MR sequence types
Example figure from our publication:
Sample segmentations by all methods on AMOS. (a) Pancreas slice, where all methods do not fully segment the region. (b) 3D renderings for another subject. Arrows indicate segmentation errors in the region of corresponding color.
For further information, please see a previous version of an arXiv paper:
Ciausu C, Krishnaswamy D, Billot B, Pieper S, Kikinis R, Fedorov A. Towards Automatic Abdominal MRI Organ Segmentation: Leveraging Synthesized Data Generated From CT Labels. arXiv preprint arXiv:2403.15609. 2024 Mar 22. https://arxiv.org/abs/2403.15609
The following table holds the regions that each method segments:
| Benchmarked Method | # Regions | Regions Segmented |
|---|---|---|
| ABDSynth | 33 | liver, spleen, kidneys, stomach, duodenum, pancreas, gallbladder, small bowel, colon, adrenal glands, sacrum, hip bones, glutei maximi, glutei medii, glutei minimi, autochthonous, iliopsoases, and the individual vertebrae from L5 to T10 |
| MRSegmentator | 40 | liver, spleen, kidneys, stomach, duodenum, pancreas, gallbladder, small bowel, colon, adrenal glands, sacrum, hip bones, glutei maximi, glutei medii, glutei minimi, autochthonous, iliopsoases, lung (left and right), heart, aorta, inferior vena cava, portal vein and splenic vein, iliac arteries, iliac veins, esophagus, urinary bladder, spine, femurs |
| MRISegmentator-Abdomen | 62 | liver, spleen, kidneys, stomach, duodenum, pancreas, gallbladder, small bowel, colon, adrenal glands, sacrum, hip bones, glutei maximi, glutei medii, autochthonous, iliopsoases, lung (left and right), aorta, inferior vena cava, portal vein and splenic vein, iliac arteries, iliac veins, esophagus, ribs 4–12 (left and right), vertebrae L1–L5 and T7–T12 |
| TotalSegmentator MRI | 59 | liver, spleen, kidneys, stomach, duodenum, pancreas, gallbladder, small bowel, colon, adrenal glands, hip bones, glutei maximi, glutei medii, glutei minimi, autochthonous, iliopsoases, lung (left and right), heart, aorta, inferior vena cava, portal vein and splenic vein, iliac arteries, iliac veins, esophagus, urinary bladder, prostate, sacrum, vertebrae, intervertebral discs, spinal cord, humeri, scapulae, claviculae, femurs, brain |