Extraction

This capsule uses a mix-and-match approach to combine Cellpose, Suite2p, and CaImAn for cell detection and signal extraction.

Input

All parameters are passed to extraction.py using python extraction.py [parameters].
They are defined in parse_args() at the top of extraction.py using argparse.
The most important one is --input-dir which should point to a directory that includes an hdf5 file with a 'data' 3D array of the registered (or, in the case of multi-plane imaging, decrosstalked) movie.

Output

The main output is the extraction.h5 file.
For QC, the figures detected_ROIs.png and detected_ROIs_withIDs.png display contours of the detected ROIs over a selection of summary images.
If the capsule is run with the --contour_video argument, a 10x sped-up video is also created, showing ROI contours overlaid on the raw data, along with ROI activity and the residual.
The extraction.h5 file contains three groups: traces, rois, and cellpose, with the latter only present if Cellpose is used for initialization (--init). The exact contents further depend on the used --neuropil method, with the bolded fields below always present.

traces:

• corrected: Neuropil-corrected fluorescence traces
• roi: Neuropil-corrupted raw fluorescence traces of each ROI
• neuropil: Neuropil fluorescence traces
• neuropil_rcoef: Coefficients used for neuropil correction
• raw_neuropil_rcoef_mutualinfo: Raw coefficients for neuropil correction (minimizing mutual information between corrected and neuropil trace)
• skew: Skewness of neuropil-corrected fluorescence traces
• std: Standard deviation of neuropil-corrected fluorescence traces

rois:

• data: Values of the (weighted) ROIs
• coords: Index locations of every data value
• shape: Shape of the ROI array, # neurons x height of FOV (in pixels) x width of FOV (in pixels)
• neuropil_coords: Index locations of neuropil masks
• aspect_ratio, compact, footprint, med, mrs, mrs0, npix, npix_norm, npix_norm_no_crop, npix_soma, overlap, radius, solidity, soma_crop: Additional ROI stats, see suite2p documentation

cellpose:

• masks: Labeled image, where 0=no masks; 1,2,…=mask labels.
• flows_in_hsv: XY flow in HSV 0-255
• flows: XY flows at each pixel
• cellprob: Cell probability at each pixel
• final_locations: Final pixel locations after Euler integration
• styles: Style vector summarizing each image

Additional datasets at the root of extraction.h5 are:

• iscell: Specifies whether a ROI is a cell, first column is 0/1, and second column is probability that the ROI is a cell based on the default classifier
• meanImg: The mean image of the movie over time
• maxImg: The maximum intensity projection of the movie over time