Accelerates Orthopedic Clinical Decisions Through Automated Knee Adduction Moment Predictions

This technology trains an algorithm to accurately predict knee adduction moments during walking. It uses kinematic and ground reaction force data, iteratively refining and validating the predictive model across different subjects to minimize error and ensure high accuracy.

Background

Biomechanics and gait analysis play a vital role in understanding human movement, particularly for diagnosing and managing degenerative musculoskeletal conditions like knee osteoarthritis. A key metric in this field is the knee adduction moment, which serves as a primary indicator of the mechanical load placed on the inner compartment of the knee joint during walking. Accurately monitoring these rotational forces is essential for clinicians and researchers to assess disease progression, design effective rehabilitation protocols, and evaluate orthopedic interventions. Consequently, there is a strong clinical need for reliable methods to continuously assess knee joint loading during patients’ daily activities.

Currently, measuring these joint forces requires specialized motion capture laboratories equipped with multi-camera systems and floor-embedded force plates. While highly accurate, this setup is expensive, time-consuming, and restricts patient assessment to artificial environments that do not reflect natural walking patterns. Recent efforts to overcome these spatial limitations use portable wearable sensors paired with predictive algorithms to estimate joint loads outside the lab. However, these algorithmic approaches often struggle to generalize across diverse patient populations. Because human movement varies significantly between individuals, existing predictive models frequently suffer from high error rates and poor accuracy when applied to new subjects.

Technology

This biomechanical technology provides a specialized method for training algorithms to predict knee adduction moments (KAM), which measure the inward-turning forces placed on the knee joint during walking. The process begins by capturing a person’s physical movement data alongside two distinct sets of ground reaction forces as they walk over pressure-sensitive floor plates. Combining the movement metrics with the first set of force data allows the system to establish a baseline reference for knee stress. The system then iteratively builds a predictive model using data from multiple individuals, continuously adjusting its internal parameters to minimize errors between its predictions and the baseline metrics.

What differentiates this approach is its rigorous, subject-isolated validation process that ensures high predictive accuracy across diverse populations. During each training cycle, the algorithm calibrates its calculations using data from a broader group while intentionally excluding one specific individual. Once prediction errors are minimized for the main group, the system generates an accuracy score by testing the model against the excluded person’s baseline data. Repeating this cycle across different subjects ensures the algorithm generalizes well to new people, resulting in a highly reliable biomechanical assessment tool.

Benefits

  • Accurately predicts knee adduction moments (KAM) by continuously adjusting internal parameters to minimize prediction errors
  • Improves model generalizability across different individuals by validating predictions against reference data from excluded subjects
  • Quantifies model reliability by generating an objective accuracy score for each test subject
  • Enables potentially simpler future biomechanical assessments by training the algorithm to predict complex joint moments from kinematic and ground reaction force inputs

Applications

  • Clinical gait analysis software
  • Knee osteoarthritis progression monitoring
  • Athletic injury prevention assessment
  • Orthopedic rehabilitation progress tracking
  • Custom orthotics and brace design

Additional info

This method trains a predictive algorithm for knee adduction moments (KAM) using kinematic and dual-set ground reaction force (GRF) data. Reference KAM is derived from kinematics and baseline GRF. The system iteratively optimizes internal model parameters to minimize error relative to reference data across training subjects, then evaluates performance by scoring predictive accuracy against a held-out subject’s reference KAM data.