Deep Learning of EEG Data in the NeuCube Brain-Inspired Spiking Neural Network Architecture for a Better Understanding of Depression

• Published in: Neural Information Processing Vol. 11955; pp. 195 - 206
• Main Authors: Shah, Dhvani, Wang, Grace Y., Doborjeh, Maryam, Doborjeh, Zohreh, Kasabov, Nikola
• Format: Book Chapter
• Language: English
• Published: Switzerland Springer International Publishing AG 2019; Springer International Publishing
• Series: Lecture Notes in Computer Science
• Subjects: Depression; Electroencephalogram (EEG); NeuCube; Spiking Neural Networks (SNN)
• ISBN: 3030367177; 9783030367176
• ISSN: 0302-9743; 1611-3349
• DOI: 10.1007/978-3-030-36718-3_17

In the recent years, machine learning and deep learning techniques are being applied on brain data to study mental health. The activation of neurons in these models is static and continuous-valued. However, a biological neuron processes the information in the form of discrete spikes based on the spike time and the firing rate. Understanding brain activities is vital to understand the mechanisms underlying mental health. Spiking Neural Networks are offering a computational modelling solution to understand complex dynamic brain processes related to mental disorders, including depression. The objective of this research is modeling and visualizing brain activity of people experiencing symptoms of depression using the SNN NeuCube architecture. Resting EEG data was collected from 22 participants and further divided into groups as healthy and mild-depressed. NeuCube models have been developed along with the connections across different brain regions using Synaptic Time Dependent plasticity (STDP) learning rule for healthy and depressed individuals. This unsupervised learning revealed some distinguishable patterns in the models related to the frontal, central and parietal areas of the depressed versus the control subjects that suggests potential markers for early depression prediction. Traditional machine learning techniques, including MLP methods have been also employed for classification and prediction tasks on the same data, but with lower accuracy and fewer new information gained.