Explainable AI based neurofeedback in a CNN-BCI for motor imagery classification

• Main Author: Ghosh, Amitrajit
• Other Authors: Guan Cuntai
• Format: Final Year Project
• Language: English
• Published: Nanyang Technological University 2024
• Subjects: Computer and Information Science; Engineering
• Online Access: https://hdl.handle.net/10356/175251
• Institution: Nanyang Technological University

The application of machine learning to brain-computer interfaces has recently garnered great interest due to the significant advancements in the areas of neural networks and deep learning. With the use of state-of-the-art convolutional neural networks, it has become possible to achieve accuracies of up to 84.19% for subject-independent left/right motor-imagery models using deep convolutional neural networks. The aim of this project is to interpret the classification of deep convolutional neural networks and study interrelations between classification performance and network encodings with neurophysiological features underlying motor imagery. To achieve that goal, this project investigated the relationship between saliency maps derived from the output of a deep convolutional neural network and a subject’s event-related synchronization/desynchronization (ERDS) values. For some electrode channels located on the motor cortex, strong positive correlations were found between ERDS and saliency values measured from those channels. These findings provide insights on the strategies employed by subjects when performing left/right motor imagery tasks. This information can be utilised in the future to further optimise the deep convolutional neural network to improve the performance of left/right motor imagery classification.