EEG-Based Depression Detection Using a Hybrid LSTM-Spiking Neural Network

Abstract

Depression is a very common mental health problem that highly influences people's quality of life and poses significant challenges to clinical diagnosis. Electroencephalogram (EEG) signals, which show the brain's electrical activity, have emerged as an important tool for detecting depression. This paper proposes a novel hybrid model that combines Long Short-Term Memory (LSTM) networks with Spiking Neural Networks (SNNs) to effectively classify depression from EEG signals. The LSTM component of the model is particularly good at retaining temporal relationships within EEG data, handling long sequences and identifying important features associated with depressive states. The SNN component, which models the spiking dynamics of biological neurons, also improves the model's capacity for handling the spatiotemporal patterns present in EEG signals and offering a more biologically plausible way of representing the data. Through the amalgamation of the strengths of the LSTM and SNN architectures, the hybrid model obtains a more stable and accurate classification of depression.

The experimental method employed here is the use of the synaptic time-dependent plasticity (STDP) learning rule in a 3-dimensional brain-template organized SNN model. Experimental results show that the model proposed in this paper outperforms classical machine learning algorithms and single architecture-based models in sensitivity, specificity, and classification accuracy. The approach has excellent classification accuracy with average performance rates at 98%. This far surpasses other top deep learning methods. This method represents a significant leap forward in automated depression detection with EEG signals, with potential real-time clinical utility and personalized treatment planning.