0
$\begingroup$

I have seen number of approaches that have been introduced in the literature for spike detection including, amplitude thresholding method, Nonlinear Energy Operator method, and template matching method. However, If I am to use the amplitude thresholding method, how can I identify a suitable threshold value for my dataset? I have seen several works using this threshold to be X times the standard deviation above the mean noise level. But this X seems to be different from one study to another. For example, commonly used values are 5, 5.5, and 6.

My question is how can I determine a suitable X for my dataset? My dataset is a set of MEA recordings taken from organoids in which I do not have any prior knowledge about neuron firing.

$\endgroup$
0
$\begingroup$

For threshold-based spike detection, the threshold defines your performance in the detection vs false alarm performance. That is to say, a lower threshold will improve your probability of detecting (smaller) spikes but eventually, some of the noise may be considered a spike, increasing the probability of a false alarm, detecing a spike where there is none. A common term for showing this trade-off is ROC.

Therefore, it is highly application-dependent how to choose the threshold. If you have a specific false alarm rate that is tolerable, you can use the noise distribution to determine the threshold that gives you the desired false alarm rate (refered to as the constant false alarm rate (CFAR) approach). Same is true if you have a tolerable missed detection rate (àla I need to find at least 99.99% of the spikes).

If not, you should consider the threshold a variable and do your evaluations as a function of the threshold. This allows you to pick a threshold that optimizes your final figure of merit.

| improve this answer | |
$\endgroup$
  • $\begingroup$ Thanks but In my case, since I do not have any ground truth with me, is what your possible? $\endgroup$ – Tyesh Jul 18 at 3:58
  • $\begingroup$ Well if you know nothing then either you just pick one, which is always an arbitrary choice, or you just try all of them until you are satisfied with your results. $\endgroup$ – Florian Jul 19 at 7:58

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.