Electroencephalogram (EEG) Classification using a bio-inspired Deep Oscillatory Neural Network

V. Srinivasa Chakravarthy,Sayan Ghosh,C Vigneswaran,NR Rohan

DOI: https://doi.org/10.1101/2024.05.24.595714

2024-05-30

Abstract:In this paper, we propose two models of oscillatory neural networks the Deep Oscillatory Neural Network (DONN) and a convolutional variation of it named Oscillatory Convolutional Neural Network (OCNN) and apply the models to a variety of problems involving the classification and prediction of Electroencephalogram (EEG) signals. Deep neural networks applied to signal processing problems will have to incorporate various architectural features to remember the history of the input signals e.g., loops between the layers, gated neurons, and tapped delay lines. But real brains have rich dynamics expressed in terms of frequency bands like alpha, beta, gamma, delta, etc. To incorporate this aspect of brain dynamics in a Recurrent Neural Network (RNN) we propose to use nonlinear oscillators as dynamic neuron models in the hidden layers. The two oscillatory deep neural networks proposed are applied to the following EEG classification and prediction problems: Prediction of nearby EEG channels, classification of single-channel EEG data (healthy vs. epileptic, different stages of sleep stage classification), and multi-channel EEG data (Epileptic vs. Normal, Left vs. right-hand Motor imagery movement, and healthy vs. Claustrophobic EEG).

Bioengineering

What problem does this paper attempt to address?

This paper mainly discusses the use of the biologically inspired Deep Oscillatory Neural Network (DONN) and its convolutional variant (OCNN) for the classification and prediction of electroencephalogram (EEG) signals. Traditional deep neural networks require the memory of input signal history when processing signals, while the brain's dynamics are reflected in different frequency bands such as alpha, beta, gamma, etc. Therefore, the researchers propose to use nonlinear oscillators as the dynamic neural model in the hidden layer of the RNN to simulate this dynamic characteristic of the brain.

The paper applies the DONN and OCNN models to various EEG classification and prediction tasks, including predicting neighboring EEG channels, single-channel EEG data (healthy vs epilepsy, sleep stage classification), and multi-channel EEG data (epilepsy vs normal, left vs right hand motor imagery, healthy vs claustrophobia). These models demonstrate performance comparable to or even better than RNNs, and they are capable of handling time series and video data.

The experimental datasets include various types of EEG data, such as IITD epilepsy data, BONN dataset, CHB-MIT dataset, motor imagery data, and sleep EEG data. The paper also introduces the architecture of the DONN model, where oscillator neurons adopt the Hopf oscillator model, and the network has a feedforward structure but can process sequential data.

Through these models, the researchers demonstrate the potential in the field of EEG technology, with wide-ranging applications from clinical settings to biobotics. The results section of the paper provides in-depth discussions on the performance of the models in various EEG processing tasks.