A Raspberry Pi-Based Traumatic Brain Injury Detection System for Single-Channel Electroencephalogram

Area of Science:

• Biomedical Engineering
• Computational Neuroscience
• Medical Diagnostics

Background:

• Traumatic Brain Injury (TBI) is a significant cause of mortality and long-term disability.
• Current TBI diagnostic methods are often subjective or require specialized clinical resources.
• Advancements in portable computing and machine learning offer opportunities for improved TBI detection.

Purpose of the Study:

• To develop a compact, portable system for real-time Traumatic Brain Injury (TBI) detection using machine learning.
• To automate the analysis of single-channel Electroencephalogram (EEG) signals for identifying mild TBI (mTBI).
• To enable early TBI detection in field settings without specialized medical equipment.

Main Methods:

• A Raspberry Pi-based system was designed for real-time EEG data acquisition and processing.
• An Analog to Digital Converter (ADC) was used to digitize the EEG signal.
• Convolutional Neural Networks (CNN) and XGBoost models were employed for signal classification and mTBI detection.

Main Results:

• The system achieved a peak classification accuracy exceeding 90% for distinguishing TBI from control conditions.
• Real-time classification was accomplished in under 1 second for epoch lengths of 16-64 seconds.
• The system demonstrated versatility by successfully operating with multiple machine learning models.

Conclusions:

• The developed portable system enables efficient, automated mTBI detection from single-channel EEG signals.
• This technology can facilitate early TBI diagnosis in diverse environments, supporting both clinical applications and research.
• The system's architecture paves the way for connected, real-time health monitoring systems for TBI-related conditions.