[Eeglablist] How many electrodes tolerated for interpolation?

[봉수현]

Dear Prof. Makeig and Chang,

While reading this interesting thread, I came across your comment and had a question.
Up to now, I have typically interpolated bad channels at the sensor level before performing source localization.

If I have understood correctly, do you recommend rejecting bad channels without interpolation and then projecting the remaining channels as they are, without restoring the rejected channels (for example, when 64 channels are reduced to 60, multiplying the 60-channel time-series data by the leadfield matrix)?

Since the leadfield matrix can simply be computed using the reduced set of channel locations, there does not seem to be any major procedural issue. However, I would like to ask whether you would expect any substantial difference between results obtained with and without interpolation in this case. My own intuition is that there would not be a large difference.

Best regards,

Bong

-----Original Message-----
From:  "장진원 via eeglablist" <eeglablist at sccn.ucsd.edu>
To:      <smakeig at gmail.com>; 
Cc:      <eeglablist at sccn.ucsd.edu>;  "fernandez luis" <isabelyluis2007 at me.com>; 
Sent:  2025-11-19 (수) 09:19:43 (UTC+09:00)
Subject: Re: [Eeglablist] How many electrodes tolerated for interpolation?

Dear Prof. Makeig,

I appreciate your suggestion. Is source-level EEG analysis more reliable
than scalp-level analysis? I wonder whether source-level EEG analysis (such
as dipole fitting) achieves high test-retest reliability and replicability
over different datasets, especially in clinical populations (depression vs
healthy controls, as an example)? I am not sure whether source-level
analysis is applicable in clinical research.

Best Regards,
Jinwon Chang

2025년 11월 18일 (화) 오후 1:19, Scott Makeig <smakeig at gmail.com>님이 작성:

> I would question the value of channel interpolation - except for making
> smooth pictures of particular scalp distributions. Individual
> electrode-signal-difference scalp channels are inherently vague measures
> that sum electrical activities generated in many unrelated parts of cortex.
> Scalp channel signals are not more deserving of attention than individual
> radio-frequency channels in an fMRI system ...
>
>  I do understand that (at least in the US) clinicians need to work in
> terms that insurance companies will reimburse. But cortical source-resolved
> EEG recording and analysis is technically quite feasible, as our work at
> SCCN over the last 30 years has abundantly demonstrated -- and allows more
> exact interpretation (and high statistical certainty) than scalp-level data
> interpretation.
>
> Scott Makeig
>
> On Tue, Nov 18, 2025 at 9:41 AM 장진원 via eeglablist <
> eeglablist at sccn.ucsd.edu> wrote:
>
>> Thank for your kind reply. I understand it.
>>
>> Best Regards,
>> Jinwon
>>
>> On Mon, Nov 17, 2025 at 7:23 PM fernandez luis via eeglablist <
>> eeglablist at sccn.ucsd.edu> wrote:
>>
>> > > Hi Jinwon,
>> > >
>> > > Accurate EEG channel interpolation is methodologically feasible
>> across a
>> > broad spectrum of montage densities, including low-density
>> configurations
>> > such as 19- and 32-channel systems, intermediate-density arrays such as
>> 64
>> > channels, and high-density systems such as 128 channels. Importantly,
>> the
>> > methodological validity of interpolation does not primarily depend on
>> the
>> > absolute number of electrodes, but on the availability of non-artifacted
>> > neighboring electrodes with adequate spatial distribution surrounding
>> the
>> > channel to be reconstructed.
>> > >
>> > >
>> > > EEG interpolation is a spatial estimation procedure in which the
>> signal
>> > of an artifacted electrode is reconstructed using mathematically
>> weighted
>> > contributions from the nearest clean surrounding electrodes.
>> Accordingly,
>> > if the spatially adjacent electrodes are also artifacted, interpolation
>> > becomes unreliable or methodologically inappropriate, because the
>> > reconstruction process would be driven by distorted input data—violating
>> > key assumptions underlying spherical spline interpolation,
>> inverse-distance
>> > weighting, and other spatial estimation algorithms.
>> > >
>> > >
>> > > In low-density montages (e.g., 19 or 32 channels), interpolation
>> remains
>> > technically feasible; however, the reduced spatial sampling inherently
>> > limits the anatomical precision and spatial granularity of the
>> > reconstructed signal. Nevertheless, interpolation in these systems can
>> > yield clinically acceptable results as long as the electrodes used as
>> > sources for reconstruction are clean, stable, and sufficiently
>> distributed
>> > around the artifacted location.
>> > >
>> > >
>> > > Intermediate-density systems such as 64-channel EEG offer improved
>> > spatial resolution that allows more accurate reconstruction of missing
>> > channels due to enhanced scalp coverage. High-density montages,
>> > particularly 128-channel EEG systems, provide dense and homogeneous
>> spatial
>> > sampling, minimizing interpolation error and generating reconstructions
>> > that are more physiologically plausible and quantitatively reliable.
>> This
>> > level of spatial resolution is advantageous for applications requiring
>> > high-fidelity scalp mapping, microstate analysis, connectivity
>> estimation,
>> > and source localization.
>> > >
>> > >
>> > > Despite differences in resolution across montage densities, a
>> > fundamental methodological requirement remains invariant: interpolation
>> > must be performed exclusively using clean, non-artifacted surrounding
>> > electrodes. Reconstruction based on artifacted neighbors compromises the
>> > physiological validity of the estimated signal and undermines the
>> > mathematical assumptions intrinsic to spatial interpolation algorithms.
>> > >
>> > >
>> > > Technical Comparison: Conventional vs. High-Density EEG Interpolation
>> > >
>> > >
>> > > 1. Interpolation in conventional EEG (19–32 channels)
>> > >
>> > >
>> > > Low-density EEG systems rely on sparse spatial sampling, which imposes
>> > several methodological constraints:
>> > >
>> > >
>> > > • Wide inter-electrode spacing
>> > >
>> > > • Higher vulnerability to local contamination
>> > >
>> > > • Reduced capacity to capture rapid spatial changes
>> > >
>> > > • Acceptable but limited reliability
>> > >
>> > >
>> > > 2. Interpolation in high-density EEG (64–128 channels)
>> > >
>> > >
>> > > High-density EEG (HD-EEG) significantly enhances the reliability of
>> > interpolation due to:
>> > >
>> > >
>> > > • Dense and homogeneous spatial sampling
>> > >
>> > > • Robustness to isolated corrupted channels
>> > >
>> > > • Improved modeling of spatial gradients
>> > >
>> > > • Near-physiological reconstruction in 128-channel systems
>> > >
>> > > Best
>> > > Luis Fernandez, MSc
>> > > Clinical Neuropsychologist
>> > >
>> >
>> > > El 17 nov 2025, a las 20:41, 장진원 via eeglablist <
>> > eeglablist at sccn.ucsd.edu> escribió:
>> > >
>> > > Hi all,
>> > >
>> > > I'm a clinical psychiatrist, so I am not really familiar with
>> engineering
>> > > concept of interpolation. I believe in high-density setting (128
>> channel)
>> > > interpolation of a few channels are acceptable, but what if more than
>> 10
>> > > bad channels in 64 channel setting? Is it tolerable or detrimental for
>> > > maintenance of true signal?
>> > >
>> > > Best regards,
>> > > Jinwon Chang
>> > > _______________________________________________
>> > > To unsubscribe, send an empty email to
>> > eeglablist-unsubscribe at sccn.ucsd.edu or visit
>> > https://sccn.ucsd.edu/mailman/listinfo/eeglablist    .
>> >
>> > _______________________________________________
>> > To unsubscribe, send an empty email to
>> > eeglablist-unsubscribe at sccn.ucsd.edu or visit
>> > https://sccn.ucsd.edu/mailman/listinfo/eeglablist   .
>> _______________________________________________
>> To unsubscribe, send an empty email to
>> eeglablist-unsubscribe at sccn.ucsd.edu or visit
>> https://sccn.ucsd.edu/mailman/listinfo/eeglablist  .
>
>
>
> --
> Scott Makeig, Research Scientist and Director, Swartz Center for
> Computational Neuroscience, Institute for Neural Computation, University of
> California San Diego, La Jolla CA 92093-0559, http://sccn.ucsd.edu/~scott  
>
_______________________________________________
To unsubscribe, send an empty email to eeglablist-unsubscribe at sccn.ucsd.edu or visit https://sccn.ucsd.edu/mailman/listinfo/eeglablist .