larget latency for P300

[PeterLiu]

Hi,

We're doing research on P300-based BCI with BCI2000 and we found that the latency of the peak is around 600ms after stimulus onset.
We collected data from 12 subjects, the latency were all similar (around 600ms). We used Neuroscan 4.5 for data acquisition. For the 12 subjects, the application is a 3 by 3 matrix for spatial navigation with stimulus duration 80ms and ISI 160ms. We also did some dry runs with standard 6 by 6 P3 speller, the results are still similar (with latency around 600ms). We tried both using BCI2000's offline analysis and the signal processing tools we developed. Is this because of communication/screen display delay or there are any other possible explanation? Thank a lot!

Yichuan

[mellinger]

Hi,

there is nothing wrong with your setup. Rather, apparent latency is larger than actual latency due to a number of factors:

*The BCI2000 StimulusCode state will be set when a block of data is being processed, not when it is acquired. This means that stimulus onset will appear earlier by about one block length (SampleBlockSize parameter, given in samples).

*In addition, the Neuroscan amplifier will accumulate data into an internal buffer before sending them to BCI2000. The length of this buffer will add to the observed latency.

*BCI2000 records the point in time when video memory was altered, not the point in time when the stimulus actually appears on screen. There may be an additional difference of up to two frames (33ms) between the two time points.

From your description of the setup, it seems that the Neuroscan amplifier sends data in packets of 80ms, and that BCI2000 sample block size matches this block duration. Altogether, you would thus have an additional time lag of between 170 and 190ms. You would thus find the peak between 410 and 430 ms, which is fine for a typical P300 response (300ms + half peak width).

If you are interested in exactly measuring absolute latency of the observed P300 response, you should run your experiment with a light sensor attached to one of the matrix elements on the screen, and the light sensor's output being recorded in an EEG channel of the amplifier. This allows you to measure the additional offset caused by the effects described above.

Regards,
Juergen

[PeterLiu]

This explains, Thanks a lot!