Deep learning approach to predict pain progression in knee osteoarthritis

Area of Science:

• Orthopedics
• Radiology
• Artificial Intelligence

Background:

• Knee osteoarthritis (OA) is a leading cause of pain and disability.
• Predicting OA pain progression is crucial for timely intervention.
• Current risk assessment models have limitations in accuracy.

Purpose of the Study:

• To develop and evaluate deep learning (DL) models for predicting knee OA pain progression.
• To compare the performance of DL models against traditional risk assessment methods.

Main Methods:

• Retrospective analysis of the Osteoarthritis Initiative cohort (9348 knees).
• Development of a DL model using baseline knee radiographs.
• Development of a traditional model using demographic, clinical, and radiographic factors.
• Creation of a combined model integrating DL and traditional factors.
• Evaluation using Area Under the Curve (AUC) on a hold-out testing dataset.

Main Results:

• The DL model achieved an AUC of 0.770, outperforming the traditional model (AUC 0.692).
• The combined model demonstrated the highest performance with an AUC of 0.807.
• DL models showed significantly higher diagnostic performance for predicting pain progression.

Conclusions:

• Deep learning models utilizing baseline knee radiographs offer superior diagnostic performance for predicting knee OA pain progression.
• The combined DL and traditional factor model provides the most accurate prediction.
• These findings support the use of DL in knee OA management.