Fall risk classification for people with lower extremity amputations using random forests and smartphone sensor...

Area of Science:

• Biomedical Engineering
• Rehabilitation Science
• Wearable Technology

Background:

• Fall-risk classification is crucial for preventative programs, yet models often exclude lower limb amputees.
• Lower limb amputees face a higher fall risk than older adults, necessitating specific predictive tools.
• Current fall-risk assessment lacks dedicated models for amputee populations.

Purpose of the Study:

• To develop and validate a predictive fall-risk classification model for lower limb amputees.
• To utilize smartphone sensor data during a 6-minute walk test (6MWT) for fall-risk assessment.
• To identify key gait features indicative of fall risk in amputee individuals.

Main Methods:

• Collected data from 89 lower limb amputees performing a 6MWT with a smartphone on the pelvis.
• Extracted 248 gait features from accelerometer and gyroscope data, segmenting steps into turns and straight walking.
• Employed feature selection techniques and a random forest classifier to build the fall-risk model.

Main Results:

• The optimal model utilized turn-step data, selected by Correlation-based Feature Selection (CFS), with 500 random forest trees.
• Achieved classification metrics: 81.3% accuracy, 57.2% sensitivity, 94.9% specificity, 0.587 MCC, and 0.81 F1 score.
• Performance comparable to existing clinical fall-risk assessment tools.

Conclusions:

• A smartphone-based gait analysis method shows promise for classifying fall risk in lower limb amputees.
• The developed classifier may serve as a viable tool for clinical practice and personalized fall prevention.
• Further validation could integrate this technology into routine rehabilitation for amputee populations.