Amputee Fall Risk Classification Using Machine Learning and Smartphone Sensor Data from 2-Minute and 6-Minute Walk...

Area of Science:

• Biomedical Engineering
• Rehabilitation Science
• Artificial Intelligence

Background:

• The 6-minute walk test (6MWT) assesses mobility, but AI can extract deeper insights like fall risk.
• The 2-minute walk test (2MWT) is an alternative for individuals with limited mobility, including lower limb amputees.
• Automated foot strike (FS) detection offers potential for enhanced gait analysis.

Purpose of the Study:

• To investigate automated foot strike (FS) detection using AI for the 2-minute walk test (2MWT).
• To evaluate the efficacy of AI-derived gait parameters from the 2MWT for fall risk classification in lower limb amputees.

Main Methods:

• A Long Short-Term Memory (LSTM) model was trained for automated foot strike (FS) detection using retrospective 6-minute walk test (6MWT) data.
• The LSTM model was trained on full 6MWT data and then on the initial 2 minutes for 2MWT application.
• A random forest model used step-based features from automated FS for fall risk classification.

Main Results:

• The 6-minute LSTM model achieved high accuracy (99.2%) for FS detection.
• The 2-minute LSTM model demonstrated good accuracy (98.0%) but lower sensitivity (65.0%) for FS detection.
• Fall risk classification using 2-minute automated FS data had 76.3% accuracy but failed to adequately identify individuals with a recent fall history.

Conclusions:

• Automated foot strike (FS) detection is feasible for both 6MWT and 2MWT using LSTM models in lower limb amputees.
• While FS detection is accurate, features derived from a 2-minute test duration are insufficient for reliable fall risk classification in this population.