Effect of noise in recordings on potential change in the waveform. More realiable to remove or to filter?

[Kayv-cmb]

Hi Spikeinterface,

I am mostly looking for advice and this may not be the place to ask, but i assumed you would know the answer to my interrogation.
I have a strange noise in my neuropixels recordings where the channels get really noisy for somewhere bw 0.1-2 seconds and this noise is completely synchronus and periodic, and I would say during those time the waveform are not in a good shape .
I have a put a screenshot of a subset of channel where this is happening (Top part unfiltered)
As you can see in the bottom part of my plot after filtering(bandpass,phase shift and cmr) the noise is gone as it should with CMR.

However i am afraid that those noisy part could have change the waveform of the spike and that my spike waveform could be affected after filtering.

Would that affect my spike sorting?

I would prefer a shorter but realiable recording, than a complete dodgy one. I know I can slice the recordings, but is there a function in spikeinterface that can help me caracterise and found those noisy part, so that I can remove it afterward from the recording ?

I would kindly like your point of view on the effect of noise on the spike waveform and if I should get the trouble of completly removing those part ?

Thank you in advance, My question is not an issue in spikeinterface, so I thought it would be more appropriate to ask here instead.

Kayv_cmb

[alejoe91]

Hi @Kayv-cmb

Sorry for our late reply here! We're still getting used to the Discussion tab.

In general, any processing will change the waveforms, including filtering. The question is more about what kind of processing will result in a better detectability and discoverability of spikes. In the example you showed, CMR seems very beneficial!

You can take a look at the remove_artifacts function. Here, you have to provide a list of intervals to mask/remove priot to spike sorting. We're also working on implementing some automatic methods for artifact detection, but it's still WIP.

[Kayv-cmb]

Hi @alejoe91,

No worries, I was mostly looking for your input on change in waveform I didn't expect a fast answer , I already created a function that would detect those period and remove them, In the process I remove some spike that are not in those period but it's kinda hard to target only this part because I have no idea where they are.

I kind of expect that the waveform will change with processing but the change would be similar across the recording so I expect that the clustering would work correctly? (I am unsure)

However the change cause by this noise which I think is electrical, like a surcharge in the device or change in the impedance of the electrode, would change the waveform ? And that change would only be localize on those periods with this noise. Is it possible to quantify those change in waveform if I don't know which cluster in the sorting are the one from noise and not noise that need to be merge.

More globally is there a way to track waveform that would change because of either external change (like any processing) or any internal change (like modification of the membrane potential) in the neurons it belong that is not happening during the entire recording?

I don't know if my question are clear, no worries if you don't have any answer.

[alejoe91]

yes, as long as waveform changes are consistent across chanels, it should make clustering more precise!

I don't think there is a straightforward way to track down changes due to internal variability and external noise sources, since we don't know the internal variability a priori. There might be some options to look into that, but I'm not sure you need to go down this complex path