Posthoc Interpretability of Neural Responses by Grouping Subject Motor Imagery Skills Using CNN-Based Connectivity

Area of Science:

• Neuroscience
• Biomedical Engineering
• Machine Learning

Background:

• Motor Imagery (MI) is crucial for Brain-Computer Interfaces (BCI), but performance varies due to user skill and EEG signal limitations.
• Decoding neural responses from EEG for MI-BCI is challenging due to signal non-stationarity and poor spatial resolution.
• A significant portion of users struggle with MI tasks, leading to underperforming BCI systems and 'BCI-Inefficiency'.

Purpose of the Study:

• To develop a method for early identification of subjects with poor motor imagery skills in BCI training.
• To propose a Convolutional Neural Network (CNN)-based framework to analyze neural responses and improve MI-BCI performance.
• To address inter/intra-subject variability in MI EEG data for more robust BCI control.

Main Methods:

• Utilized a CNN framework to analyze high-dimensional dynamical data from EEG signals during Motor Imagery tasks.
• Extracted functional connectivity from spatiotemporal class activation maps using a novel kernel-based cross-spectral distribution estimator.
• Employed subject clustering based on classifier accuracy to identify common and discriminative motor skill patterns.

Main Results:

• Achieved an average accuracy enhancement of 10% compared to the baseline EEGNet approach on a bi-class dataset.
• Reduced the proportion of 'poor skill' subjects from 40% to 20% through the proposed method.
• Demonstrated the framework's ability to interpret neural responses in subjects with deficient MI skills and high EEG variability.

Conclusions:

• The proposed CNN-based framework effectively identifies subjects with poor motor imagery skills early in BCI training.
• The method improves MI-BCI accuracy and reduces the number of underperforming users.
• This approach offers a way to understand and potentially improve BCI performance even for individuals with significant skill deficits.