0% accuracy in P300 BCPy2000 Application

[corbitv]

Hello,

I've posted a few times concerning a P300 memory game application that my research team created with BCPy2000. The game is simple at face value; participants select a "card" from a grid (4 columns, 3 rows), after the flashing sequence is finished the card "flips" to reveal an image, another flashing sequence commences during which the participant chooses another card, that card flips, and if a match is made they both stay showing and if a match isn't made they are both changed back to the "card back."

After a few hang ups, we have the program running fine, except that we have 0% accuracy. Even when using a parameter created on a speller matrix of the same size, none of the computer's choices are correct. The classification results reported do not seem right either; according to one of our programmers, the two selections (highest row score and highest column score) should be very different from all of the other scores. All of the scores we are seeing are fairly similar (see attached example output). The program still chooses the highest two values, though. So, it seems that the issue is somewhere between the EEG signals from the brain and the computer's interpretation of them. Spelling on the modified speller (3x4) with that same parameter worked perfectly.

Even more curiously, when the game was first created (August 2011), it had fairly high accuracy (only missing a few choices). It was not used again until April 2012, at which point it was accurate for the first few turns of use, and then became inaccurate for every use since then.


Unfortunately, I did not program the game and do not know how to program. However, I am in the process of learning and more than willing to learn how to work with these programs. I just don't know where to even start to look to try and figure out to fix this problem. I read on the BCPy2000 site that the program is kind of broken up into different sections (Process, Transition, Phases, etc) ... does one of these sections deal with the the issue we seem to be having? I know it may be difficult to help someone who doesn't actually know how to program, but any possible advice that could lead me in the right direction would be greatly, greatly appreciated. Thank you very much!

[corbitv]

As a followup to that post:

We suspect that the problem may be in the synchronization of the EEG P300 signals with the flashing of the application. As I said, the classification results do not appear to be right, as if the program isn't actually registering the P300 signals that occur with the flashes. Does seem like a plausible explanation?

Since it appears that the accuracy decreased over time, we were thinking that it might be a timing issue that compounds.

Hopefully the further specificity of my question will aid in any possible advice.

Thank you.

[boulay]

Do you have a photosensor that you can hook up as an EEG input? Setup your source and signal processing to only use a single channel then plug the photosensor into that channel. Load the data offline and look at the relative timing of the targetcode state change and the response. I don't have much time to get into this right now but I saw you haven't had a reply yet so I thought I'd give you something to try.