A machine learning-based diagnostic model associated with knee osteoarthritis severity.

Area of Science:

• Biomechanics
• Medical Engineering
• Data Science

Background:

• Knee osteoarthritis (KOA) significantly impacts patient-reported outcomes, including pain and reduced gait function.
• Objective measures are needed to accurately assess KOA severity and guide treatment.
• Patient-reported outcome measures (PROMs) like the WOMAC index are commonly used but can be subjective.

Purpose of the Study:

• To identify gait features associated with KOA severity using PROMs.
• To develop machine learning models for estimating KOA severity based on gait data.
• To explore the clinical utility of gait analysis in KOA evaluation.

Main Methods:

• Collected gait data from 375 volunteers with varying KOA grades.
• Utilized Western Ontario and McMaster Universities Osteoarthritis Index (WOMAC) scores to define KOA severity.
• Extracted 1087 gait features, applying ANOVA and t-tests for feature selection (p < 0.0001).
• Developed linear regression and random forest models to predict WOMAC scores.

Main Results:

• Identified 43 significant gait features across multiple joints and spatiotemporal parameters.
• Selected features strongly correlated with WOMAC subscales: physical function (41 features), pain (10 features), and stiffness (16 features).
• Machine learning models achieved a predictive correlation of 0.741 for KOA severity.

Conclusions:

• Gait analysis, incorporating machine learning, provides an objective method for estimating KOA severity.
• The identified gait features and developed model can aid in personalized rehabilitation and progress monitoring.
• This approach offers a promising tool for clinical application in KOA assessment.