Large-scale separation of neural spikes from stimulation artifacts code

This repo contains (matlab) sample code for doing spike sorting with stimulation artifacts.

• See Example.m for the full execution of the code, along with details.
• Art0.mat contains the initial artifact that is used to find the model hyperparameteres. This Artifact is a three dimensional tensor (Art0=Art0(j,e,t)) with dimensions E=512 (electrodes), T=55 (time samples), J=34 (number of different electrical stimuli). This initial estimate was built using equation (7) in the paper. Also, hyperparameters based on this estimate are computed according to equation (8).
• Data.mat contains the actual data where spikes are to be found. Specifically

1. The fourth dimensional tensor TracesAll(j,n_j,t,e) of recordings over array for all trials (n_j=51 is the number of trials) at all amplitudes of stimulation.
2. The EI of 24 neurons (figure 1 here and in the paper), represented as a the cell array templates. Each of these EIs is represented as a matrix V(e,t) which states how a spike is recorded in each electrode.
3. The list of stimulation amplitudes, listAmps
4. Stimulating electrode indexes stimElecs(here, 404).
5. breakpoints contains the indexes of stimuli at which breakpoints (Sudden changes in the artifact measured in the stimulating electrode) occurred

• After running Example.m two outputs should appear:
  • A sample of the spatial profile of the artifact, as figure 3 in the paper.
  • The infered responses (activation curves) of each of the neurons of the array, and their EIs (another representation of the first figure shown here)

Array support

This code is based on a 512-array, the one depicted below(each circle represents and electrode, and the number its index)

• The structure Array in /Utils, Array.m contains array-specific information (e.g. positions, spacing, etc). Specifics for a distinct, 519-array (see below) are also shown there, to illustrate that the algorithm depends on the array through those parameters, and therefore, extensions are straightforward as long as assumptions hold.