Profiling a Raspberry Pi-Based Motor Imagery Classification to Facilitate At-Home BCI for Children with Disabilities

Area of Science:

• Neuroscience
• Computer Science
• Biomedical Engineering

Background:

• Brain-Computer Interface (BCI) systems face challenges in real-world adoption due to computational limitations and lengthy calibration.
• Moving BCI technology from lab settings to home environments is crucial for individuals with severe physical disabilities.

Purpose of the Study:

• To evaluate the feasibility of using a Raspberry Pi 4 (RPi4) for at-home BCI applications.
• To analyze the performance of various machine learning classifiers on the RPi4 for BCI tasks.
• To identify computational bottlenecks and resource usage of BCI processes on the RPi4.

Main Methods:

• Profiling ten standard classifiers, including Riemannian Geometry (RG) and Artificial Neural Network (ANN), on an RPi4.
• Comparing RPi4 performance metrics (training time, inference time, memory usage) against Desktop and MacBook computations.
• Assessing accuracy, recall, precision, and F1-score for evaluated neural network models.

Main Results:

• RPi4 demonstrated comparable performance metrics (84.3% accuracy, recall, F1-score; 84.7% precision) for neural network models.
• Profiling revealed RPi4's computational characteristics: 1.74s training time, 0.405s inference time, 1154.9 MiB peak memory, and 405.2 MiB incremental memory usage.
• Identified computational bottlenecks, suggesting careful consideration for high-resource demanding classifiers like ANN.

Conclusions:

• The RPi4 is a potentially viable device for developing low-cost, scalable at-home BCI systems.
• Implementation of resource-intensive classifiers like ANN on RPi4 requires careful planning to manage cost and complexity.
• Findings support the potential for more accessible and practical BCI solutions for individuals with disabilities.