Utility of Cognitive Neural Features for Predicting Mental Health Behaviors

Abstract

Cognitive dysfunction underlies common mental health behavioral symptoms including depression, anxiety, inattention, and hyperactivity. In this study of 97 healthy adults, we aimed to classify healthy vs. mild-to-moderate self-reported symptoms of each disorder using cognitive neural markers measured with an electroencephalography (EEG). We analyzed source-reconstructed EEG data for event-related spectral perturbations in the theta, alpha, and beta frequency bands in five tasks, a selective attention and response inhibition task, a visuospatial working memory task, a Flanker interference processing task, and an emotion interference task. From the cortical source activation features, we derived augmented features involving co-activations between any two sources. Logistic regression on the augmented feature set, but not the original feature set, predicted the presence of psychiatric symptoms, particularly for anxiety and inattention with >80% sensitivity and specificity. We also computed current flow closeness and betweenness centralities to identify the “hub” source signal predictors. We found that the Flanker interference processing task was the most useful for assessing the connectivity hubs in general, followed by the inhibitory control go-nogo paradigm. Overall, these interpretable machine learning analyses suggest that EEG biomarkers collected on a rapid suite of cognitive assessments may have utility in classifying diverse self-reported mental health symptoms.

Keywords: EEG; anxiety; depression; hyperactivity; inattention; logistic regression; machine learning; mental health; source localization.

Figure 1

Cognitive assessments delivered on the BrainE platform (adapted from [12]). (A) BrainE assessment dashboard with the wireless EEG recording setup. (B) The selective attention and response inhibition tasks differ only in the frequency of targets; sparse 33% targets appear in the Selective Attention block and frequent 67% targets appear in the Response Inhibition block. (C) Working memory task with perceptually thresholded stimuli. (D) Flanker interference processing task; flanking fish may either face the same direction as the middle fish in congruent trials, or the opposite direction in incongruent trials. (E) Emotion interference task presents neutral, happy, sad, or angry faces superimposed on an arrow.

Figure A1

The EEG processing pipeline followed in Balasubramani et al., (2021). The source-localized data from that study were used for our present analysis.

Figure A2

The figure shows the stimulus-locked ERSP results from Balasubramani et al., (2021). The source−localized data from that study were used for our present analysis.

Figure A3

The figure shows the statistically corrected source-localization results from Balasubramani et al., (2021). The heatmap colors represent Bonferroni-corrected significant theta and alpha band event-related synchronization (ERS) and beta band event-related desynchronization (ERD) during stimulus encoding relative to baseline for each of the five cognitive tasks, and for the global cognitive task average. This source-localized data were used for our present analysis.

Figure 2

Various statistical measures added as new features.

Figure 3

Prediction accuracy of logistic regression for symptom scores for top-k augmented features in chi-square statistic. As number of features (horizontal axis) increases, the accuracy initially increases rapidly and then reaches a plateau around 40,000 features (black vertical dash-line).

Figure 4

Comparison of performance of logistic regression with and without modification of the original dataset. Bar plots show comparison of prediction performance, sensitivity (left column), and specificity (right column) of 4 mental health symptoms (i) anxiety, (ii) depression, (iii) inattention, and (iv) hyperactivity, by logistic regression applied to different datasets.Original: original dataset (first row); OS: over-sampled (SMOTE and adding Gaussian noise) dataset (second row); DA + FS + OS: dataset that underwent all three processes (third row), data augmentation (increase of feature), feature selection (reduction of feature) based on chi-square statistic, and oversampling (SMOTE and adding Gaussian noise).

Figure 5

Topographical plot showing “strength” of ROI connections of relevance to mental health disorders, anxiety (upper left), depression (upper right), inattention (bottom left), and hyperactivity (bottom right). Size of each vertex corresponds to sum of chi-square statistic of the vertex. An edge is created when the sum of chi-square statistic of the two vertices that share the edge exceeds threshold: 100. The thickness of all edges is the same.

Figure 6

Frequency bands with highest cumulative chi-square scores across ROIs. (left) Topographical plot of ROIs whose edges are colored to represent the frequency band with the largest cumulative chi-square score. (right) The cumulative chi-square value of each frequency band in the top 5 ROIs.

Figure 6

Frequency bands with highest cumulative chi-square scores across ROIs. (left) Topographical plot of ROIs whose edges are colored to represent the frequency band with the largest cumulative chi-square score. (right) The cumulative chi-square value of each frequency band in the top 5 ROIs.

Figure 7

Tasks with highest cumulative chi-square scores across ROIs. (left) Topographical plot of ROIs whose edges are colored to represent the cognitive task with the largest cumulative chi-square score. (right) The cumulative chi-square value of each of the 5 tasks in the top 5 ROIs. Go Green I is a go-nogo selective-attention task, Go Green II is a go-nogo response inhibition task, Lost Star is a working memory task, Middle Fish is an interference processing task, and Face Off is an emotional interference processing task.

Figure 7

Tasks with highest cumulative chi-square scores across ROIs. (left) Topographical plot of ROIs whose edges are colored to represent the cognitive task with the largest cumulative chi-square score. (right) The cumulative chi-square value of each of the 5 tasks in the top 5 ROIs. Go Green I is a go-nogo selective-attention task, Go Green II is a go-nogo response inhibition task, Lost Star is a working memory task, Middle Fish is an interference processing task, and Face Off is an emotional interference processing task.