Associating Knee Osteoarthritis Progression with Temporal-Regional Graph Convolutional Network Analysis on MR Images

Area of Science:

• Artificial intelligence in medical imaging
• Graph convolutional networks for disease progression
• Osteoarthritis research

Background:

• Artificial intelligence (AI) shows potential for assessing knee osteoarthritis (OA) progression on MRI scans.
• Current AI methods face challenges in accuracy and interpretability for OA progression assessment.

Purpose of the Study:

• To introduce a temporal-regional graph convolutional network (TRGCN) for analyzing knee OA progression on MR images.
• To investigate the association between knee OA progression status and TRGCN outcomes.

Main Methods:

• A retrospective study utilized MR images from 194 OA progressors and 406 controls from the OA Initiative.
• Anatomical subregions (cartilage, bone, meniscus, fat pad) were segmented to form compartment-based graphs for the TRGCN model.
• The TRGCN model incorporated both regional and temporal information, with performance compared against clinical variables, radiologist scores, radiomics, and a Densenet-169 CNN.

Main Results:

• The composite TRGCN model achieved superior performance with AUCs of 0.841 (DESS) and 0.856 (IW) in the testing cohort.
• Interpretability analysis revealed cartilage as a key structure (42%-45%) and highlighted the tibiofemoral joint's (TFJ) greater importance over the patellofemoral joint (PFJ).
• Temporal analysis showed dynamic importance score changes in compartments over time, particularly differentiating TFJ and PFJ progression.

Conclusions:

• The composite TRGCN demonstrates superior discriminative ability for knee OA progression compared to other methods.
• The TRGCN model effectively captures temporal and regional information, offering interpretable insights into knee OA progression.
• This AI approach provides a valuable tool for identifying and understanding knee OA progression.