A hybrid spiking neural framework for depression recognition with EEG signals

There are two principle approaches for depression modeling and simulation in diagnosis and analysis: neuroscience-based biomedical modeling and data-driven machine learning modeling. Integrating them into one network may provide complementary modeling capabilities and efficiency. In human brain systems, hybrid global-local learning principles are broadly applicable in the modification of neural synaptic weights. Here, we present a brain-inspired hybrid global-local learning model by introducing error-driven global learning and neuroscience-oriented local learning with a differential spiking model incorporating neuronal dynamics and synaptic plasticity. By hybrid optimization, it can predict trends of Electroencephalogram (EEG) waveforms from depressive patients and healthy controls recognize whether the subjects are depressive, and analyze what physiological elements are related. We demonstrate the advantages of this model compared to state-of-the-art(SOTA) methods in depression recognition. It achieves both higher recognition performance and complementary physiological evidence. Due to the performance in simulation trends of EEG waves, this model has the potential to solve few-shot problems in clinical depression diagnosis.