Patterns of Shared Variation in Knee Ultrasound for Osteoarthritis: A Machine Learning Approach.

Area of Science:

• Orthopedics
• Radiology
• Machine Learning

Background:

• Knee osteoarthritis (KOA) is a degenerative joint disease with varied clinical presentations.
• Understanding distinct KOA phenotypes is crucial for targeted treatment strategies.

Purpose of the Study:

• To identify novel knee osteoarthritis (KOA) phenotypes using demographic, clinical, symptomatic, and ultrasound (US) data.
• Apply a machine learning approach to uncover shared variation patterns in KOA.

Main Methods:

• Utilized data from the Johnston County Health Study, including demographics, clinical assessments, radiographs, and knee US.
• Employed the Angle-based Joint and Individual Variation Explained (AJIVE) algorithm to analyze shared and individual data variations.
• Focused on shared structures between US features and non-US clinical data in 861 participants, including 335 with radiographic KOA (rKOA).

Main Results:

• AJIVE identified two significant components of shared variation (SC1 and SC2).
• SC1 linked osteophytes and cartilage damage (US) with higher BMI, age, and worse symptoms.
• SC2 correlated effusion and synovitis (US) with better function and lower BMI, suggesting an inflammatory subtype.

Conclusions:

• Two distinct KOA phenotypes were identified, potentially representing osteophyte/cartilage damage and inflammatory subtypes.
• These findings suggest clinically feasible phenotypes that warrant further investigation and validation.