Streamlined Gesture and Arm Motion Analysis via Direct EIT Voltage Classification

Area of Science:

• Biomedical Engineering
• Machine Learning
• Signal Processing

Background:

• Electrical Impedance Tomography (EIT) is a noninvasive imaging method using electrical data for human body imaging.
• EIT has applications in medical diagnostics and human-computer interaction, but image reconstruction is computationally intensive.
• This limits EIT's real-time application potential due to power and processing demands.

Purpose of the Study:

• To develop a faster and more accurate method for gesture classification using EIT data.
• To bypass the computationally expensive image reconstruction phase in EIT.
• To enhance the efficiency and reduce power consumption for real-time EIT applications.

Main Methods:

• A novel Statistical Analysis, Information Theory, and Data-Driven (SID) pipeline was developed.
• The SID pipeline analyzes raw EIT voltage data using statistical techniques, information theory, and feature ranking.
• Machine learning models, specifically XGBoost, were used for classification on reduced feature sets.

Main Results:

• The SID pipeline significantly reduced feature sets for gesture recognition and arm flexion/extension tasks.
• XGBoost classification achieved 90.38% accuracy for gesture recognition and 100.00% for arm flexion/extension.
• The method demonstrated high accuracy and computational efficiency by bypassing EIT image reconstruction.

Conclusions:

• The proposed SID pipeline combined with machine learning offers a highly accurate and computationally efficient approach for EIT-based gesture recognition.
• This method significantly reduces power consumption and processing requirements, making it suitable for real-time applications.
• This work highlights the potential of direct machine learning application on raw EIT data for enhanced human-computer interaction and diagnostics.