An Integrated Computer Vision and Force Sensing Framework for Automated Fugl-Meyer Hand-Related Assessment Using...

Area of Science:

• Biomedical Engineering
• Rehabilitation Technology
• Artificial Intelligence in Healthcare

Background:

• The Fugl-Meyer Assessment (FMA) is crucial for evaluating post-stroke motor function but suffers from subjective scoring.
• Objective measurement tools are needed to enhance the reliability and consistency of FMA evaluations.

Purpose of the Study:

• To develop and validate a portable system using computer vision and force sensing for objective FMA scoring.
• To pre-train an artificial neural network (ANN) on healthy subject data for automated FMA scoring and assess its transferability to stroke patient data.

Main Methods:

• A portable multi-camera and force-sensing system was designed to capture hand position, joint angles, and grasp strength.
• Data from healthy subjects performing FMA tasks were used to pre-train eight different ANN architectures.
• The optimal ANN model was tested on data from stroke patients, comparing its performance to clinical therapist assessments.

Main Results:

• The optimal ANN model achieved 98% accuracy on healthy subject data and 85% accuracy on stroke patient data without fine-tuning.
• Autoencoder-based feature extraction and Long Short-Term Memory (LSTM) methods improved accuracy and captured temporal motion dynamics.
• The study demonstrated successful knowledge transfer from healthy subject data to stroke patient data, mitigating data collection challenges.

Conclusions:

• The developed system effectively captures hand motions and grasp strength for objective FMA scoring.
• Pre-trained ANN models using healthy subject data can be effectively transferred to clinical stroke patient data.
• This research provides a foundation for advanced transfer learning strategies to improve automated stroke motor function assessment.