The automated feature extraction capabilities of deep learning classifiers have promoted their broader application to EEG analysis. In contrast to earlier machine learning studies that used extracted features and traditional explainability approaches, explainability for classifiers trained on raw data is particularly challenging. As such, studies have begun to present methods that provide insight into the spectral features learned by deep learning classifiers trained on raw EEG. These approaches have two key shortcomings. (1) They involve perturbation, which can create out-of-distribution samples that cause inaccurate explanations. (2) They are global, not local. Local explainability approaches can be used to examine how demographic and clinical variables affected the patterns learned by the classifier. In our study, we present a novel local spectral explainability approach. We apply it to a convolutional neural network trained for automated sleep stage classification. We apply layer-wise relevance propagation to identify the relative importance of the features in the raw EEG and subsequently examine the frequency domain of the explanations to determine the importance of each canonical frequency band locally and globally. We then perform a statistical analysis to determine whether age and sex affected the patterns learned by the classifier for each frequency band and sleep stage. Results showed that δ, β, and γ were the overall most important frequency bands. In addition, age and sex significantly affected the patterns learned by the classifier for most sleep stages and frequency bands. Our study presents a novel spectral explainability approach that could substantially increase the level of insight into classifiers trained on raw EEG.

Background: Electroconvulsive Therapy (ECT) is one of the most effective treatments for major depressive disorder (DEP). There is recently increasing attention to evaluate ECT's effect on resting-state functional magnetic resonance imaging (rs-fMRI). This study aims to compare rs-fMRI of DEP patients with healthy participants, investigate whether dynamic functional network connectivity network (dFNC) estimated from rs-fMRI predicts the ECT outcome, and explore the effect of ECT on brain network states. Method: Resting-state fMRI data were collected from 119 patients with depression or DEP (76 females), and 61 Healthy (HC) participants (34 females) with an age mean of 52.25 (N=180) years old. The pre-ECT and post-ECT Hamilton Depression Rating Scale (HDRS) were 25.59+/-6.14 and 11.48+/-9.07, respectively. Twenty-four independent components from default mode (DMN) and cognitive control network (CCN) were extracted using group-independent component analysis from pre-ECT and post-ECT rs-fMRI. Then, the sliding window approach was used to estimate the pre-and post-ECT dFNC of each participant. Next, k-means clustering was separately applied to pre-ECT dFNC and post-ECT dFNC to assess three distinct states from each participant. We calculated the amount of time each individual spends in each state, called occupancy rate or OCR. Next, we compared OCR values between HC and DEP participants. We also calculated the partial correlation between pre-ECT OCRs and HDRS change while controlling for age, gender, number of treatment, and site. Finally, we evaluated the effectiveness of ECT by comparing pre-and post-ECT OCR of DEP and HC participants. Results: The main findings include: 1) DEP patients had significantly lower OCR values than the HC group in a state, where connectivity between CCN and DMN was relatively higher than other states (corrected p= 0.015), 2) Pre-ECT OCR of state, with more negative connectivity between CCN and DMN components, predicted the HDRS changes (R=0.23 corrected p=0.03). This means that those DEP patients who spend less time in this state showed more HDRS change, and 3) The post-ECT OCR analysis suggested that ECT increased the amount of time DEP patients spend in state 2 (corrected p=0.03). Finally, we found ECT increases the total traveled distance in DEP. Conclusion: Our finding suggests that dFNC features, estimated from CCN and DMN, show promise as a predictive biomarker of the ECT outcome of DEP patients. Also, this study identified a possible underlying mechanism associated with the ECT effect in DEP patients.