Unsupervised machine-learning classification of electrophysiologically active electrodes during human cognitive task performance

Abstract

Identification of active electrodes that record task-relevant neurophysiological activity is needed for clinical and industrial applications as well as for investigating brain functions. We developed an unsupervised, fully automated approach to classify active electrodes showing event-related intracranial EEG (iEEG) responses from 115 patients performing a free recall verbal memory task. Our approach employed new interpretable metrics that quantify spectral characteristics of the normalized iEEG signal based on power-in-band and synchrony measures. Unsupervised clustering of the metrics identified distinct sets of active electrodes across different subjects. In the total population of 11,869 electrodes, our method achieved 97% sensitivity and 92.9% specificity with the most efficient metric. We validated our results with anatomical localization revealing significantly greater distribution of active electrodes in brain regions that support verbal memory processing. We propose our machine-learning framework for objective and efficient classification and interpretation of electrophysiological signals of brain activities supporting memory and cognition.

Independent dynamics of slow, intermediate, and fast intracranial EEG spectral activities during human memory formation

“…We first identified a subset of electrodes from cortical areas activated during memory encoding. For this purpose, spectral power-in-band in six non-overlapping frequency ranges (low theta: 2-4 Hz, high theta: 5-9 Hz, alpha: 10-15 Hz, beta: 16-25 Hz, low gamma: 25-55 Hz, high gamma: 65-115 Hz) was used as features for automated classification of active electrodes (Saboo et al 2019) independently in each band (Fig. 1).…”

“…Active electrodes were classified independently in six frequency bands of the iEEG spectrum using normalized estimates of the power change (z-score transform). The classification procedure used an unsupervised method based on the Gaussian Mixture Model (Saboo et al 2019). Notice induced activity across all six frequency bands in the example spectrogram plotted from trial-averaged activity from one electrode localized in the occipital cortex.…”

“…Each trial was normalized separately. Log normalization was implemented in order to correct for the 1/f-power law effect of lower frequencies on estimating power in the higher frequency bands, and the z-score transformation was performed to provide a normalized scale of power change for comparison of signals from different electrodes, sessions and participants (Kucewicz et al 2019; Saboo et al 2019; Kucewicz et al 2017). This z-score normalization produces positive and negative power changes relative to the mean power within any one word epoch instead of using a “baseline” period.…”

“…1. Electrodes were then classified into “active” and “inactive” in each specific band using our unsupervised clustering method based on Gaussian Mixture Modeling (GMM) to identify those that showed task-induced power changes (Saboo et al 2019). We chose a GMM-based method to ensure that different numbers of electrodes would be allowed for the clusters of active electrodes and the inactive electrodes.…”

Optimal Adaptive Electrode Selection to Maximize Simultaneously Recorded Neuron Yield

“…This work is related to channel selection methods developed for EEG [9, 19] and ECoG [26] applications. These existing methods select channels to reduce dimensionality as a preprocessing step for decoding or to identify physiologically active channels.…”