This is a method to perform the cell segmentaiton step in caclium imaging analysis, which uses the temporal information from caclium imaging videos in form of correlations, and combines a deep learning model with an instance segmentation algorithm.

"DISCo: Deep learning, Instance Segmentation, and Correlations for cell segmentation in calcium imaging", E. Kirschbaum, A. Bailoni, F. A. Hamprecht, arXiv preprint arXiv:1908.07957, 2019. [pdf]

• Python 3.6 (or later): we recommend installing it with Anaconda
• PyTorch 1.0 (or later)
• GASP
• inferno 0.3 (or later)

1. Download or clone this repository
2. Install GASP as described here
3. Get inferno as described here
4. Download the neurofinder training and test data from here
5. Extract the neurofinder data into HDF5 files:
   • create for each neurofinder video a HDF5 file with a dataset named 'video' containing the video with shape time x X x Y
   • create a file named BF_labels.h5 containing the foreground-background labels for each video
   • create a file summary_images.h5 containing the mean intensity projection for each video
   • create a file gt_segmentations.h5 containing the instance labels for each video

Example:
python run.py -p ../neurofinder_videos/ -m disco -gpu 1