using slow wave activity from a P3 speller in BCI2000

Forum for discussion on different signal processing algorithms
Locked
Brendan
Posts: 12
Joined: 18 Aug 2003, 18:42

using slow wave activity from a P3 speller in BCI2000

Post by Brendan » 20 May 2004, 08:40

Hello,

I have an idea that may improve P3 BCIs.

I recently spent a full weekend demonstrating a P3 BCI at a public expo called NextFest. I was the experimenter, BCI2000 operator, and the subject. While doing 50+ consecutive runs with a 6 x 6 Donchin matrix for 7 hours, I had a lot of time to muse about the mental strategies I used, consciously or otherwise. I noticed that I unconsciously anticipated the mean inter target interval (ITI). That is, once my target flashed, I expected it would flash again within about 5-7 flashes. It seemed to me that, when the next target flash didn't happen for a long time, I felt a very strong sense of anticipation. And similarly, if my target flashed twice in a row, I was a little surprised. Both of these events will happen sometimes by chance.

This may be reflected in slow wave activity. As I argued previously, it is very unfortunate that designers of P300 BCIs only look at P300s. Other activity, including early components and late slow save activity, also varies for targets vs. nontargets and this can be used to inspire a pattern recognition system. I saw a lot of slow wave activity in my NextFest data, and have seen it in other subjects. In fact, research on the CNV would argue that there should be a difference in the slow wave preceding a target flash after a long ITI vs. a short ITI.

Does this effect matter? I believe this could be determined using existing BCI2000 data. Here is how:

1) determine each time that stimuluscode (the row or column flash) equals the target. This will happen twice for each set of 12 flashes, once for a row flash and once for a column flash. This is very easy; in fact, BCI2000 already does this automatically, it is an essential part of any P3 speller.

2) For each such instance, determine how many flashes have elapsed since the previous target flash. This number will vary between zero and 20 in a 6 x 6 matrix.

3) Average the resulting ERPs. I actually think we would be interested in both the ERPs to that target (when the expectation of a succeeding target drops to its minimum) and to the preceding flash (when the expectation of a succeeding target may be high or low). My hypothesis is that you will see a greater slow wave if the ITI is large vs. when it is small. I could be wrong, but thanks to the wonders of science, we can find out.

4) This would be harder to implement for a P3 speller in which the target is not known a priori - that is, any realworld use of the speller. It would have to work as follows. For each of the 12 flashes, create an "expectation table" of slow waves. That is, you're saying, "If this flash is the target, then I would expect the slow waves from that flash and the preceding flash to look like this.... if that flash is the target, then I would instead expect this pattern of slow wave activity." Then, compare the data to that expectation table and use this to help the discrimination.



Gerv - am I correct that steps 1-3 could be done fairly easily with existing data? Would #4 be a horrific hack, or is it something a good programmer could do in a reasonable time frame?
-Brendan

gschalk
Posts: 615
Joined: 28 Jan 2003, 12:37

P3 and SW ...

Post by gschalk » 21 May 2004, 12:12

Hey Brendan,

This is a good idea. I don't think anybody has looked at it. My gut feeling is that, even if there is differences in the slow waves, single-trial variability might be too high to really add much information. However, your hypothesis and steps 1-3 could easily be tested. In the event that there is great information in the slow wave activity, I am sure one could also do step 4) in a reasonable time frame. I just think that - on a single trial/flash basis - variability will be too high to make this a useful avenue.

Gerv

Locked

Who is online

Users browsing this forum: No registered users and 3 guests