Dsort spike sorter Installation
Dsort requires CUDA in order to run. This is because the normalized cross correlation routine is coded in CUDA (no cpu version is provided with Dsort).
What to do to run Dsort?
- Install matlab R2018a
- Install cuda 9.0 https://developer.nvidia.com/cuda-90-download-archive.
- Install visual studio 2013 https://visualstudio.microsoft.com/vs/older-downloads/
- Download Dsort
- In Matlab go to Dsort folder then subfolder CUDA, and run the mexGPUall.m, this will compile the CUDA part of Dsort. For more information about CUDA and MATLAB check https://se.mathworks.com/help/distcomp/run-mex-functions-containing-cuda-code.html
Use Dsort
- Open the config_file.m adapt the code to your settings, the settings are described in the comments of the config.file
- Run the Master file
Output The results are saved in a mat file called rez.mat. The saved variables are:
- ops contains all the parameter provided by user from the config file
- xc x coordinate of the probe
- yc y coordiante of the probe
- xcoord idem
- ycoord idem
- connected active channels (1) inactive channels(0)
- Wrot Whitenning matrix
- temp contain number of batch
- st3 detected spikes using the NCC without post-processing
- dWU Template used for detection (Number of sampling point by Number of Channel by number of template matrix)
- W Temporal component (dWU = WdU, here W= W*d)
- U Spatial component
- Weight Weight used for normalization (Nrank matrix)
- st Detected spikes after post-processing
- Chan Best channels for each template (dWU)
- Chan_score Score of each template using the 1st component of matrix U
- score NCC results between template used for normalization
- Merge_cluster Summary of which template has been merged, how many spikes, best channels
- Best_Channel_M Value of each detected spikes at its peak on all channels
- template Recalculated template
- M_template Merged template
- M_std_template Standard deviation of each merged template
- PC Projection of each spikes onto its merged_template and best channels
- PETH Auto-correlogram of each cluster
Post-processor The user has the possibility to run a post-processor to vizualize and clean the cluster found by Dsort. To do so, run the file post_master.
This displays for each cluster 4 figures:
- Projection of each spikes onto its merged_template and best channels (determined via the matrix U)
- The histogram of NCC score
- The spike amplitude for the best channels
- The average waveform with standard deviation + the PETH
The displayed prompt allow the user to choose different option:
- press 'Ok' to keep cluster
- enter '1' to delete cluster
- enter '2' to split the cluster, by default the cluster will be split in 2 using a guassian distribution on the best channel. The user can specify the number of cluster and the channel to use .
- enter '3' to clean cluster using the amplitude. The user needs to specify the amplitude threshold and the channel to use.
- enter '4' to clean the cluster using the normalized cross correlation score. The user needs to specify the NCC threshold to use.
After dealing with all the cluster the results is saved in a different mat file called "rezD.mat". The user can run the post-master as much as neccesary.