Feature Request: Support for Handling Multiple Stimulation Frequencies in LFP Data

[JasminDelVecchioDelVecchio]

Context
I am analyzing local field potential (LFP) data from two nuclei in a patient with bilateral deep brain stimulation (DBS). Each nucleus is being stimulated at a distinct frequency, with both frequencies applied simultaneously. This scenario is common in advanced DBS protocols designed to optimize therapeutic outcomes by using independent stimulation frequencies.

Problem
Existing artifact removal methods struggle with multi-frequency setups. Specifically, the challenge lies in effectively isolating and removing artifacts caused by the simultaneous application of two distinct stimulation frequencies and their harmonics and subharmonics in the same dataset. This limitation significantly affects the quality of artifact removal and downstream analyses.

Importance
Addressing this challenge would reflect realistic clinical scenarios: multi-frequency stimulation is a common configuration in DBS protocols (e.g. different stimulation frequencies for the left and right hemispheres in bilateral DBS or multi-nuclei setups).
Approaches Attempted
I have attempted the following approaches, but they were not effective:

1. Separate Filters for Each Frequency: applied two independent filters, one for each stimulation frequency, in series to the dataset.
2. Convolution of Filters: created a combined filter by convolving the two frequency-specific filters and applied this to the data.
   Feature Request
   I am requesting guidance or the implementation of tools or methods to address multi-frequency artifact removal in LFP data processing.
   Thank you for considering this feature request. This functionality would greatly benefit researchers and clinicians working with complex DBS configurations.

[theonlydvr]

Hi Jasmin,

This is a pretty interesting application and I can see how current methods don't address the artifact particularly well. I believe PARRM as is (and pretty much any window-based method) would have issues with your data because of beat frequencies arising from the interaction of your two outputs. You'd need a really large window to properly capture all the different possible combinations of the two waveforms to produce a reasonable average for removal. If your artifacts are stable, you could try using the lfpregfunction for each fundamental frequency (which you'd still need to determine with FindPeriodLFP). I do also think it might still be possible to apply the filters sequentially as long as you have accurate representations of the period for each individual frequency and a wide enough filter window.

If you have some toy data, I would be willing to give this a look to see if there are any other options.

Best,
Evan