LightFFNet: MDD Prediction on EEG Quantitative Biomarkers

Abstract

Major depressive disorder (MDD) is a global issue and every year the number of people suffering, is increasing at an alarming rate. The role of electroencephalography (EEG) in diagnosing MDD has shown a potential surge. Many studies have been carried out for designing an automated approach to the diagnosis of MDD through EEG as a primary tool of analysis. However, most of the methodologies depend on machine learning and the application of deep neural network tools. These heavily depend on the annotated EEG signals for training, which requires trained professional for data generation. In addition, the time and memory complexity of its implementations are huge. With these challenges, the article designs an approach for the detection of MDD using spectral clustering. The raw EEG is pre-processed, and then three quantitative biomarkers: band power (delta, beta, and theta band power, and three non-linear signal extracted features have been extracted from raw EEG signals. Channel-wise and hemisphere-wise analyses have been conducted to understand the correlation and reliance among the cross-hemisphere. The efficiency and effectiveness of the solution on par with the other existing design are tested and validated.