[Eeglablist] When to do the channel interpolation?

[Cedric Cannard]

Dear Nadja,

I believe the latest recommendation is to do it after ASR (if you're doing ASR) to avoid rank deficiency during the underlying PCA, and before ICA, making sure you are taking into account data rank in the input (default in EEGLAB GUI). 

See https://urldefense.com/v3/__https://www.frontiersin.org/journals/signal-processing/articles/10.3389/frsip.2023.1064138/full__;!!Mih3wA!ES0AmJX15OwJI_PSdDoXg4fRFGUpTnjazvcMHL4Apf_kSHutFHL2fmGelShMY6IDZslML1TtkJYucaVdPe7PNf6WJA$  for more details on this. 

Hope this helps,

Cedric



On Monday, 30 June 2025 at 00:37, Snegireva, N, Dr \[nsnegireva--- via eeglablist <eeglablist at sccn.ucsd.edu> wrote:

> Dear all,
> 
> May I ask for your advice on the following:
> 
> We are now pre-processing mobile EEG data from a walking study, and I am unsure at what stage to place the step of interpolating removed channels:
> 
> Some researchers performed it after ICA, e.g.:
> 
> *
> Jacobsen, Nadine Svenja Josée, Sarah Blum, Karsten Witt, and Stefan Debener. “A Walk in the Park? Characterizing Gaitrelated Artifacts in Mobile EEG Recordings.” Edited by T. SolisEscalante. European Journal of Neuroscience 54, no. 12 (December 2021): 842140. https://urldefense.com/v3/__https://doi.org/10.1111/ejn.14965__;!!Mih3wA!Eji1Zagxfk4a6quNi5eQkszCwjs_zx_ZW6by0_5e34Ge5HaDWQodZlxMKooP8BdeYEaEFcduRQJ_nOPhz93SWqmfXet7$ .
> 
> Whereas other recent relevant studies placed this step before the ICA. e.g.:
> 
> *
> Klug, M., T. Berg, and K. Gramann. “Optimizing EEG ICA Decomposition with Data Cleaning in Stationary and Mobile Experiments.” Scientific Reports 14, no. 1 (June 19, 2024): 14119. https://urldefense.com/v3/__https://doi.org/10.1038/s41598-024-64919-3__;!!Mih3wA!Eji1Zagxfk4a6quNi5eQkszCwjs_zx_ZW6by0_5e34Ge5HaDWQodZlxMKooP8BdeYEaEFcduRQJ_nOPhz93SWnJXWo3r$ .
> *
> Mustile, Magda, Dimitrios Kourtis, Martin G Edwards, Simon Ladouce, Daniele Volpe, Manuela Pilleri, Elisa Pelosin, Gemma Learmonth, David I Donaldson, and Magdalena Ietswaart. “Characterizing Neurocognitive Impairments in Parkinson’s Disease with Mobile EEG When Walking and Stepping over Obstacles.” Brain Communications 5, no. 6 (November 1, 2023): fcad326. https://urldefense.com/v3/__https://doi.org/10.1093/braincomms/fcad326__;!!Mih3wA!Eji1Zagxfk4a6quNi5eQkszCwjs_zx_ZW6by0_5e34Ge5HaDWQodZlxMKooP8BdeYEaEFcduRQJ_nOPhz93SWjckLXhT$ .
> 
> 
> I believe the caveat with the second approach (interpolating before ICA) would be that it introduces rank deficiency into the data which must be accounted for when running the decomposition, but it has the significant advantage of allowing for a spatially unbiased common average reference to be calculated beforehand. The rank deficiency can be corrected by using the 'pcakeep' option in EEGLAB to match the number of components to the true data rank. Please correct me if I am wrong...
> 
> We are not removing a great number of channels, typically its 2-4.
> Given the noisy nature of mobile data, is one approach generally recommended over the other in the community for achieving the most robust decomposition?
> Any advice would be greatly appreciated.
> Best regards,
> Nadja Snegireva
> Stellenbosch University
> 
> The integrity and confidentiality of this email are governed by these terms. Disclaimer<https://urldefense.com/v3/__https://www.sun.ac.za/emaildisclaimer/default.aspx__;!!Mih3wA!Eji1Zagxfk4a6quNi5eQkszCwjs_zx_ZW6by0_5e34Ge5HaDWQodZlxMKooP8BdeYEaEFcduRQJ_nOPhz93SWkGm7cDD$ >
> 
> Die integriteit en vertroulikheid van hierdie e-pos word deur die volgende bepalings bere?l. Vrywaringsklousule<https://urldefense.com/v3/__https://www.sun.ac.za/emaildisclaimer/default.aspx__;!!Mih3wA!Eji1Zagxfk4a6quNi5eQkszCwjs_zx_ZW6by0_5e34Ge5HaDWQodZlxMKooP8BdeYEaEFcduRQJ_nOPhz93SWkGm7cDD$ >
> 
> _______________________________________________
> To unsubscribe, send an empty email to eeglablist-unsubscribe at sccn.ucsd.edu or visit https://sccn.ucsd.edu/mailman/listinfo/eeglablist .