Predicting the Healing of Lower Extremity Fractures Using Wearable Ground Reaction Force Sensors and Machine Learning

Area of Science:

• Biomechanics
• Orthopedics
• Machine Learning in Healthcare

Background:

• Lower extremity fractures present significant challenges in healing and assessment.
• Current methods for monitoring fracture recovery are often subjective and limited.
• Wearable technology offers potential for objective, continuous patient monitoring.

Purpose of the Study:

• To investigate the efficacy of wearable gait sensors and machine learning in predicting lower extremity fracture healing.
• To develop a data-driven model for objective assessment of fracture recovery.
• To explore the potential for early detection of complications and accurate rehabilitation timeline prediction.

Main Methods:

• Retrospective analysis of gait monitoring insole data from 25 patients with closed lower extremity fractures.
• Processing of continuous underfoot loading data to extract gait metrics.
• Utilizing white-box machine learning models (decision tree, Lasso regression, logistic regression) for feature selection and prediction.
• Model evaluation using 10-fold cross-validation and leave-one-out validation.

Main Results:

• Machine learning model achieved approximately 76% accuracy, precision, recall, and F1-score in predicting fracture healing.
• Feature selection identified underfoot loading distribution patterns, especially on the medial surface, as key indicators.
• Cross-validation and leave-one-out validation demonstrated stable predictive metrics.

Conclusions:

• Integrating wearable sensors with machine learning provides objective assessment for lower extremity fracture healing.
• The developed model can accurately predict rehabilitation timelines and aid in early complication detection.
• This approach facilitates data-driven clinical decisions, potentially shortening recovery times.