Home

#Welcome to the neurofinder wiki!

###Discussion on source output format:

• 3D datasets have 3 number in “coordinates”
• Bounding box can be user specified to be larger than max / min coordinates if desired
• Neuropil can be represented as either: negative numbers in “values”, or as two sources files with matched id numbers, one neuropil, one neurons.
• Transform function will do neuropil subtraction if desired
• Transform function will do temporal unmixing between overlapping sources if desired.

Source extraction algorithms should be submitted with a transform function specified

---

Merging and blocks: Feed arbitrary list of block centers and sizes Force block to return at most one source Blocks size and merging function should be parameterized. String together feature methods, block methods and merging in a modular fashion

Evaluate metrics: Metric should treat negative values corresponding to neuropil appropriately Evaluate different portions of the algorithms separately. i.e. did you return the right number of neurons per block i.e. starting with the right centers do you find the right boundaries Evaluate on F trace. Assign quality metric to sources - high confidence, low confidence

---

For the future: Df/f to spikes Cross-day merging (do you find same neurons?) Finding axons / dendrites - block methods may work well, but need different merging function Incorporate motion correction into source extraction Estimate z-motion for timeseries

---

Ground truth & evaluation metrics

Ground truth: Manually chosen centroids + ridge finding algorithm. One for filled neurons, one for donuts.

Metrics based on centroids, pixel overlap, and F trace.

Centroid: Distance metric Binarized if under

Pixel overlap: % overlap if the values are binary Cross correlation

F trace evaluation: Correlation to ground truth F trace