Continuous Wavelet Transform Based Across Scale-Functional Connectivity Matrix for Motor Imagery EEG Classification Utilizing Modified EEG Morlet and LSTM Deep Neural Network

Brain-Computer Interface (BCI) technology stands at the forefront of interdisciplinary research, merging neuroscience, engineering, and computer science to forge direct communication channels between the human brain and external devices. BCI-based devices have tremendous applications in prosthetic device development. The challenges in real-time practical BCI implementation are due to the bulky models, inherent noises, artifacts, and complexity of motor imagery (MI) electroencephalogram (EEG) data with inter-subject and intra-subject variabilities. To overcome these challenges, the proposed algorithm introduces a modified EEG Morlet (MEM) wavelet having a better time bandwidth product leading to detailed feature extraction with capability of natural filter for artifacts and noises introduced by eye blinking and muscle movements. Further, the proposed approach utilizes Hilbert transform to extract temporal features of analytical signal, extract their common spatial patterns, calculates the continuous wavelet transform (CWT) coefficients, arranges these coefficients at different scale for each channel, calculates the cross-correlation for each scale, and observes the evolution in cross-correlation matrices at different scale with the help of customized long-short term memory (LSTM) neural network to classify MI EEG. The customized LSTM architecture had the size of 1.93 MB showing the effectiveness of methodology for MI EEG classification of embedded-based devices. The best classification accuracy achieved by MEM wavelet with instantaneous magnitude temporal feature was 83.78% and the comparative analysis with earlier state-of-the-art methods showed an improvement of 1.10% in accuracy.