What is the best way to analyze SEEG & microwire recordings with different sample rates?

[trichner]

Hi Brainstormers,

I have datasets with SEEG channels recorded at 5,000 samples/second and microwire channels recorded at 32,000 samples/second. I read in a couple places that BST doesn't typically like multiple sampling rates. And when I try to load the data (which are save in NWB format), it detects the multiple sample rates and throws an error. I'm wondering if you can suggest the best work around. Below are a few options that I'm thinking about.

(1) Upsample all signals to 32,000 samples/second. I could resave the data so that all of the data within the .nwb files are 32,000 samples/second. The files would be huge.

(2) Upsample the data on the fly within in_fread_nwb(). I read that the Spike2 reader does this (see previous discussion link)

Synchronized visualization and analysis of video and electrophysiology

Thank you! Essentially, we would like to browse elphys and video data concurrently for behavioral analysis. In our case, we would be interested in tools facilitating the post-hoc synchronization of asynchronously recorded data streams, including videos. I think this is an issue many labs come across every now and then. Example: we record EEG using dedicated sampling hardware plus behavior using one or more simple USB cameras. The two devices might have started at different time points, so ther…

(3) Process the data separately. Maybe I could edit the nwb reader to allow the user to pick the SEEG or microwire data separately. I could then process the microwire data, which will yield event times for the sorted units, and then analyze the events against the SEEG data. I'm leaning towards this appraoch.

(4) I could process the microwire data outside Brainstorm, and then just load the sorted units as events into Brainstorm. I was hoping to keep the entire pipeline within Brainstorm, but maybe that is stretching it.

If anyone has suggestions or experience to share, I'd be very grateful.

Sincerest thanks,
Tom

[tmedani]

Hi Tom,

I think 32Khz is too much for the sampling rate. Unless you have a specific reason to work in this range, I would recommend downsampling your data to the lowest common SR value (5k in your case).
In that case, it is still reasonable to use all your data and process it in BST.

Upsampling your data will not add any relevant data; you will increase the size of the data heavily, and it will also slow the performance of the software/computer.

[trichner]

Thanks for your reply. It is at 32 kHz because there are single units. I was planning to use one of the spike sorters like WaveClus.