A Scoping Review of Machine Learning Approaches for Predicting Lower Extremity Joint Contact Loads: Current Trends,...

Area of Science:

• Biomechanics
• Computational modeling
• Machine learning applications in healthcare

Background:

• Human gait analysis is crucial for assessing locomotion and musculoskeletal disorders.
• Instrumented 3D gait analysis is the gold standard, but physics-based models offer deeper insights.
• Machine learning (ML) is emerging as a viable alternative to complex simulations for clinical applications.

Purpose of the Study:

• To synthesize machine learning approaches for estimating lower extremity joint contact forces.
• To review current ML methodologies, data requirements, and validation strategies in gait analysis.
• To identify challenges and future directions for ML in predicting joint contact loads.

Main Methods:

• A systematic scoping review following PRISMA-ScR guidelines.
• Searched major scientific databases (PubMed, IEEE Xplore, Scopus, SpringerLink) from January 2014 to August 2024.
• Extracted data from 27 eligible studies on ML for lower extremity joint contact force estimation.

Main Results:

• Significant variability exists in study populations, movement types, input data, ML methods, and validation metrics.
• Small and underrepresentative datasets (particularly for females) limit model generalizability.
• Inconsistent validation and lack of open data/code impede reproducibility and comparability.

Conclusions:

• ML models demonstrate potential for accurate prediction of joint contact loads and forces.
• Future research must prioritize diverse datasets, standardized methodologies, and open science.
• Integrating physics-informed ML approaches can enhance clinical applicability of gait analysis.