Explainable Deep Learning-Based EEG Analysis for Biomarker Discovery and Its Application on Depersonalisation/derealisation Disorder

Mental health disorders are typically diagnosed based on subjective reports (e.g., through questionnaires) followed by clinical interviews to evaluate self-reported symptoms. Therefore, considering the interconnected nature of psychiatric disorders, their accurate diagnosis is a real challenge without indicators of underlying physiological dysfunction. Depersonalisation/derealisation disorder (DPD) is an example of dissociative disorder characterised mainly by persistent disembodiment, detachment from the surroundings, and feeling of emotional numbness. Its underlying neural correlates have been investigated to understand and help with a more accurate and in-time diagnosis of the disorder. However, in terms of EEG studies, which hold great importance due to their convenient and inexpensive nature, the literature has often been based on hypotheses proposed by experts in the field, meaning it requires prior knowledge of the disorder. In addition, participants labelling in research experiments are often derived from the outcome of the Cambridge Depersonalisation Scale (CDS), a subjective assessment to quantify the level of depersonalisation/derealisation. As a result, I aimed to propose a novel EEG processing pipeline based on deep neural networks to discover electrophysiological DPD biomarkers. My deep learning model requires no prior knowledge or assumption of the disorder. In addition, the structure of the proposed model targets the unreliability of CDS scores by using them as prior information only to guide the unsupervised learning task in a multi-task learning scenario. I have also presented new ways of network visualisation to investigate spectral, spatial, and temporal information derived in the learning process and have provided neuroscientific evidence supporting the reliability of my results. I have also applied the visualisation approach to a novel motor imagery BCI system called EEG-ITNet to represent the future of more robust, interpretable, and high-accuracy BCI systems. The proposed EEG analytics in this thesis could also be applied to investigate other mental disorders currently diagnosed based on clinical assessment scores.