Performance Evaluation of LSTM Models in Detection of Depression from EEG Signals

The human brain comprises millions of neurons that serve as information carriers between the brain and the body. Analysis of cognitive behavior through signals from the brain are useful in understanding various neurological disorders. Major Depressive Disorder (MDD) is a complex mental health disorder that affects individuals in varying degrees. Electroencephalography (EEG) is a useful non-invasive modality for acquiring brain signals, and its characterization and classification is facilitated through the development of automated diagnostic tools. This paper explores the use of deep learning models such as long short-term memory (LSTM), Bidirectional LSTM and Attention LSTM mechanism to identify depression patterns from EEG signals. EEG signals are acquired from depression patients and healthy controls under resting and active states for a duration of 5 minutes each from the left and right hemispheres of the brain. The ability of the LSTM models to detect depression from the EEG signals recorded from the left and right brain hemispheres is also examined, and the best EEG signal acquisition mode for accurate diagnosis is identified from the dataset. LSTM's recent advancements have enabled advanced mathematical models to consider tens of thousands of parameters, making it a promising candidate for the accurate diagnosis of depression. The study compares the accuracy of individual models and concludes that the Attention LSTM model is best suited for predicting MDD with an accuracy of 97%. The findings have implications for improving the early detection and treatment of depression.