Identification of the most important features of knee osteoarthritis structural progressors using machine learning...

Area of Science:

• Orthopedics and Biomedical Engineering
• Radiology and Medical Imaging
• Artificial Intelligence in Healthcare

Background:

• Osteoarthritis (OA) is a degenerative joint disease with significant impact on quality of life.
• Identifying early structural knee OA progressors is crucial for timely intervention.
• Current methods for predicting OA progression have limitations in accuracy and feature identification.

Purpose of the Study:

• To identify the most important baseline features for predicting structural knee osteoarthritis (OA) progression.
• To evaluate machine learning methods for classifying knee OA progressors.
• To determine the optimal classification model for early OA detection.

Main Methods:

• Utilized data from the Osteoarthritis Initiative, including 1107 baseline features and 135 quantitative magnetic resonance imaging (MRI) features.
• Defined OA progressors based on cartilage volume loss, Kellgren-Lawrence (KL) grade, and joint space narrowing (JSN) at 48 and 96 months.
• Employed six feature selection models and six classification methods, including sparse partial least square (sPLS) for feature prioritization and Multi-Layer Perceptron (MLP) for classification.

Main Results:

• Multi-Layer Perceptron (MLP) demonstrated high accuracy (AUC 0.80-0.95) for classifying progressors based on cartilage volume loss, KL grade, and JSN.
• Gradient Boosting Machine achieved an AUC of 0.70 for predicting cartilage volume loss at 48 months.
• Sparse partial least square (sPLS) identified joint space width, medial tibial plateau cartilage thickness, and JSN as top predictive features.

Conclusions:

• Baseline X-ray and MRI-derived features can effectively predict early structural knee OA progression.
• Machine learning, particularly MLP, offers a robust approach for classifying knee OA progressors.
• This study highlights the potential of integrating diverse imaging and clinical data for improved OA management.