Depression Identification Using EEG Signals via a Hybrid of LSTM and Spiking Neural Networks

Area of Science:

• Computational Neuroscience
• Artificial Intelligence
• Clinical Psychology

Background:

• Depression severity is typically assessed using subjective questionnaires like the Beck Depression Inventory (BDI).
• Quantitative depression assessment using electroencephalography (EEG) signals offers an objective alternative.
• Spiking neural networks (SNNs) are biologically realistic and suitable for modeling brain activity from EEG data.

Purpose of the Study:

• To introduce a novel framework combining SNN and LSTM for classifying depression levels from raw EEG signals.
• To model brain structures associated with different depression stages.
• To provide a quantitative, brain-inspired method for depression assessment.

Main Methods:

• Developed a hybrid SNN-LSTM framework for EEG signal processing.
• Utilized the synaptic time-dependent plasticity (STDP) learning rule within a 3D brain-template SNN.
• Classified and predicted depression outcomes, visualizing associated brain structure alterations.

Main Results:

• Achieved high classification accuracy: 98% for eyes-closed and 96% for eyes-open EEG states.
• Demonstrated superior performance compared to state-of-the-art deep learning methods.
• Provided visual representations and interpretations of brain alterations linked to depression levels.

Conclusions:

• The novel SNN-LSTM framework effectively classifies depression severity using raw EEG signals.
• This brain-inspired approach offers new insights into neurological mechanisms of depression.
• The method presents a significant advancement in objective, quantitative depression assessment.