NeuroEval

This document refers to:

(1) https://hardm.ath.cx/hard/neuro_eval.pl

and

(2) https://hardm.ath.cx/hard/turner/.

If you don't have the login/pass, you are out of luck.

(1) Is a human-ordered list of neurons recorded, almost exclusively from subcortical areas, during the surgeries while the patient was playing the game. The columns of this table are as follows:

• file,
• channel,
• unit (1 is unsorted.. have not looked too much at unsorted),
• movement buttons,
• waveform (mean +- 1std thickness)
• waveform PCA (gray = unsort, colors are a bit wrong because i was creating the figures remotely)
• ISI (I am in the process of updating this to semilog plot in the future like Shih-Chieh for looking for ISI clusters)
• Then: a inset table of analog channel v.s. information (bits) and mean shuffled information (as a control, also in bits), followed by the number of timestamps which contributed to this mutual-information estimate.
  • The entries in this sub-table are in reverse order, as follows:
    • 6 = 200ms window around target appearance.
    • 5 = 1 if abs(hand velocity) > 82% of CDF of abs(hand velocity) (e.g. 82% threshold on hand velocity).
    • 4 = abs(hand velocity)
    • 3 = hand velocity
    • 2 = target position
    • 1 = hand flex
  • The columns are highlighted in:
    • red if the max measured information > shuffled information *20,
    • orange if max measured information > shuf. info *15,
    • yellow if max meas. info > shuf. info *10. This is a heuristic, but it works.
  • The following 6 columns show the actual posterior sampling estimate, p(spike | x, lag) where x is one of the 6 analog variables above, and lag is in seconds. If you click on the pictures, you can see the full tryptic in which:
    • The top plot is the measured information over lags, assuming that each spike contributes equally to this time-stationary estimate and spikes are created by a poisson process. In this plot, the X's indicate the suffled information values, so as to get a visual sense of signifigance of the information.
    • The second plot is, as already mentioned, the p(spike | x, lag).
    • The third plot is p(spike | x, lag) smoothed over x with a polynomial of order 5. For binary-valued data, the smoothing does nothing, of course - there are only two points to sample for estimating p(x | spike) (which is later inverted through bayes' rule). For higher dimensional data this smoothing matters, as the mututal information (and hence shuffled info) is estimated between the /smoothed/ posterior and prior.

(1) links to (2), as you may have noticed. (2) and the subdirectories therein summarize ach surgery day as tables of PCA and waveform plots per file and section. Sections are created per-file based on where the electrode seems to be stable (in most of the files, we moved the electrodes; it is somewhat easy to discriminate the periods of movement by simply plotting the mean background activity, as the figures linked on the pages show.

Now, things you can use -

• For the most highly-rated neuron (page 0, leftmost p() plot), the figure indicates that the neuron is strongly inhibited from firing just before the target appears, e.g. it is a reward neuron, in some sense. This was recorded from the STN.
• The second neuron in the list is inhibited the few seconds before target appearance, and subsequently it fires, hence it is a 'target-appear' - or perhaps 'start planning to move' neuron. Note the periodic structure in the information and probability plots, which is a consequence of the patients very periodic behavior in this section. This neuron was also recorded from STN - the cortical electrodes did not work because the sping that wrapped around the cannula had, for some reason, voltage on it and this corrupted recordings until it was removed, as the lab book indicates: https://hardm.ath.cx/hard/labbook/
• The third through sixth neurons in the list (all subthalamic neurons), e.g. is of the same type - it starts firing upon target appearance. These neurons could presumably be used to gate prosthetic movements. We had a good day with the 8-ch: https://hardm.ath.cx/hard/turner/081006/PLEX081006002_s1_wf.html