No Cluster 0 - Githubissues

csn-le / wave_clus

A fast and unsupervised algorithm for spike detection and sorting using wavelets and super-paramagnetic clustering

127 stars 66 forks source link

No Cluster 0 #229

Open esbrady opened 1 year ago

esbrady commented 1 year ago

Hi Ferchaure,

I was just wondering, in instances where there is a cluster 1 but no cluster 0 (see below), does this mean that the detected units that would usually fall into cluster 0 have all grouped together to form cluster 1 instead? Is it likely that this cluster represents MUA/noise rather than SUA?

Many thanks, Erica

ferchaure commented 1 year ago

Hi Erica,

Technically cluster 0 are the spikes thar are too far away from the known templates and usually (not a real criterion) cluster 1 is the main clusters and MUA. In your case, everything is really similar to the single template found.

In my opinion: -Try with a positive threshold I saw a few positive peaks

That is MUA (or maybe just noise with a peak, try increasing a the detection threshold) and oddly fast for a neuron

esbrady commented 1 year ago

Thank you, Ferchaure! This is very helpful.

Other than noticing a large number of spikes that break the refractory limit, what are some other ways you can distinguish MUA from SUA in the wave_clus output?

ferchaure commented 1 year ago

It's complicated a few MU criteria could be:

The smallest waveform not far away from the peak... far away neurons look the same (specially when the SNR is low).
Different levels of noise in some samples that gives you the idea that there are more than one waveform there.