FOOOF aperiodic (over)estimation

[stefandukich]

Hello,

I am working with the FOOOF method using the Brainstorm implementation (FOOOF_matlab) on resting-state data (6 minutes total, 2 s epochs, 50% overlap, 0.3 Hz highpass filter).

I have noticed that the method often inaccurately estimates the spectral peaks (e.g., identifying one large peak instead of two, or completely missing a peak). To address this, I thought of estimating the periodic power manually by calculating the difference between the raw power and the estimated aperiodic component. When I do this, however, I get negative values at the "edge" frequencies (see the attached figure; averages across 70 participants).

I have tried using the default parameters as well as varying them (e.g., narrowing down the freq. span), but the overestimation of the aperiodic component persists. Also, I am using the 'fixed' aperiodic mode as this is what is typically used (which might be a questionable choice, of course).

I would be very keen to hear your opinion on this. Papers often do not plot the raw spectra versus FOOOF estimates, so I am unsure if this is a common occurrence or something specific to my data. Any advice would be appreciated.

Thanks, Stefan

[Raymundo.Cassani]

@stefandukich

I have noticed that the method often inaccurately estimates the spectral peaks (e.g., identifying one large peak instead of two, or completely missing a peak).

Have you taken a look to the model-selection addition to the spectral parametrization method?
It aims to improve the peak detection and reduce the subjective parameter choices.

https://neuroimage.usc.edu/brainstorm/Tutorials/Fooof#Model-selection_.28ms-specparam.29

@stefandukich

When I do this, however, I get negative values at the "edge" frequencies (see the attached figure; averages across 70 participants).

This is quite common, and rather than on the edge frequencies, it appears on the tails of the (Gaussian) peaks, thus it is also common to find it in the middle of the PSD in between peaks, as in this example, the green line is the raw PSD.

---

@Luc, do you want to add something else to the conversation?