Fall-Risk Classification in Amputees Using Smartphone Sensor Based Features in Turns.

Area of Science:

• Biomedical Engineering
• Rehabilitation Science
• Wearable Technology

Background:

• Falls pose a significant risk of life-altering injuries, particularly in vulnerable populations.
• Existing fall-risk classification models are effective in older adults using accelerometer data.
• The amputee population, while distinct, may also benefit from similar predictive techniques.

Purpose of the Study:

• To validate the effectiveness of fall-risk classification in individuals with lower limb amputations.
• To explore the use of smartphone-based sensor data for fall risk assessment in amputees.
• To identify optimal feature selection methods for accurate fall-risk prediction in this population.

Main Methods:

• 83 individuals with lower limb amputations participated, performing a six-minute walk test.
• An Android smartphone captured accelerometer and gyroscope data via the TOHRC Walk Test app.
• A random forest classifier was employed with feature subsets identified through three selection techniques.

Main Results:

• Correlation-based Feature Selection yielded the best results: 78.3% accuracy, 62.1% sensitivity, and a 0.51 Matthews Correlation Coefficient.
• The 'peak distinction feature' was consistently selected across all feature selection methods.
• Classification performance was comparable to studies on fall risk in elderly populations.

Conclusions:

• Smartphone-based fall-risk classification shows promise for individuals with lower extremity amputations.
• The achieved sensitivity (62.1%) and specificity (87.0%) suggest practical viability.
• Further research is recommended to enhance classification accuracy and clinical utility.