Estimation of time-to-total knee replacement surgery with multimodal modeling and artificial intelligence

Area of Science:

• Orthopedics
• Artificial Intelligence
• Medical Imaging

Background:

• Current methods for predicting time-to-total knee replacement (TKR) lack the robustness and accuracy needed for clinical decision-making.
• Accurate prediction of TKR timing is crucial for effective patient management and resource allocation in osteoarthritis care.

Purpose of the Study:

• To develop and validate an artificial intelligence (AI)-based model for predicting the time to TKR.
• To identify key features driving accelerated knee osteoarthritis progression using longitudinal data.
• To enhance the accuracy of TKR time prediction by integrating multimodal data sources.

Main Methods:

• Longitudinal data from 547 subjects (Osteoarthritis Initiative) were used for training and testing AI models for TKR prediction.
• External validation was performed using data from 518 subjects (Multi-Center Osteoarthritis Study) and 164 subjects from internal hospital data.
• Deep learning models extracted features from radiographs and MR images, which were combined with clinical and imaging assessment data for survival analysis using a Lasso Cox model and random survival forest.

Main Results:

• Predictive models using only clinical variables achieved 60.4% accuracy and a 62.9% C-index for time-to-TKR.
• Integrating deep learning features from MR images and radiographs with clinical and imaging assessment data significantly improved prediction accuracy to 73.2% (p=.001) and a C-index of 77.3%.

Conclusions:

• The developed AI model demonstrates significant potential for accurately predicting time-to-TKR surgery.
• Multimodal data fusion, incorporating deep learning features, is key to enhancing predictive accuracy.
• This AI-driven approach can assist clinicians in personalizing treatment strategies and improving patient outcomes for knee osteoarthritis.