Problems using Amplitude Artifact Detector

[nchild]

Hello, I recently started using the Amplitude Artifact Detector box in my motor tasks BCI pipeline, which has improved classification accuracy by a noticeable amount. However, there seems to be some miscommunication between the recording units of my OpenBCI Cyton+Daisy headset and the OpenVibe software, as I have to set the threshold around 90000-175000 to get good results (5-9% artifact detection w/ cutoff method). The box says the unit is mV, and obviously EEG recording is not 175V. This wouldn't really be an issue except for that the box seems to have a threshold limit of 187500, and in some of my recordings, setting the threshold to 187499 results in 100% artifact detection, then increasing it to 187500 results in 0% artifact detection, which makes the box purposeless for these recordings. This issue seems to either be from a bad recording or a software limitation, but I am still able to get classifiable results from the recordings which leads me to believe they are not entirely unusable.

I know this may be an obscure issue to run into, but would there be a way to either change the unit of my recordings so that I am using a lower value for threshold (on the order of 185 rather than 185000), or is there some way I can adjust the software to increase this limit? I am using OpenVibe Designer 3.4.0, and I have the artifact box positioned immediately after the File Reader box, before any filtration takes place. Attached is a screenshot of one of my scenarios to show how it is connected. Thank you for the help,

-Nathan

artifactdetector_issue.png (97.44 KiB) Viewed 2454 times

[kyunghowon]

Hello.
I would like to mention the potential mistakes in your scenario.

EEG amplifiers record EEG signals in either nanovolt (e-09), microvolt (e-06), millivolt (e-03), or rarely volt, depending on the company.
In general, we investigate EEG in a microvolt scale.

If the default box unit is a millivolt, you need to specify your threshold accordingly.
If necessary, you can scale data through a simple DSP box.

Another point is DC offset. I do not know which data you are loading. If you load raw data, each channel has a DC offset and power line noise, making amplitude detection meaningless. Therefore, I would recommend removing DC offset (zero-mean) or high-pass filtering/low-pass filtering (wider bandwidth than your CSP boxes) before using the amplitude detection box.

You can try displaying your signal through the signal display box, use "none" vertical scale, and then specify ylim to check your EEG signal scales.

Best regards,

[nchild]

Hello,

Thank you for the advice. I was not able to find a recording units specification of the OpenBCI equipment, but applying a bandpass filter before the amplitude artifact detection has fixed the issue. I used a 1-45hz bandpass before the artifact detection, as my CSP uses a filter bank of 4-40hz. Thank you for pointing out the DC offset and power line noise; I did not consider that, and that is probably why my units were showing up so high in the artifact detection box.
-Nathan