MR-Transformer: A Vision Transformer-based Deep Learning Model for Total Knee Replacement Prediction Using MRI.

Area of Science:

• Radiology and Medical Imaging
• Artificial Intelligence in Medicine
• Orthopedics and Musculoskeletal Diseases

Background:

• Knee osteoarthritis (OA) is a leading cause of disability, often necessitating total knee replacement (TKR).
• Accurate prediction of OA progression is crucial for timely intervention and improved patient outcomes.
• Current predictive models often lack the sophistication to fully leverage complex MRI data.

Purpose of the Study:

• To develop and evaluate MR-Transformer, a novel deep learning model for predicting knee OA progression to TKR.
• To utilize ImageNet pretraining and 3D spatial correlations for enhanced predictive accuracy.
• To compare MR-Transformer's performance against state-of-the-art deep learning models across multiple MRI sequences.

Main Methods:

• Retrospective analysis of 353 Osteoarthritis Initiative (OAI) and 270 Multicenter Osteoarthritis Study (MOST) knee MRI datasets.
• Inclusion of four MRI sequences: COR-IW-TSE, SAG-IW-TSE-FS, COR-STIR, and SAG-PD-FAT-SAT.
• Sevenfold nested cross-validation to assess MR-Transformer against TSE-Net, 3DMeT, and MRNet.

Main Results:

• MR-Transformer achieved high areas under the receiver operating characteristic curve (AUCs) across all MRI sequences (0.84-0.88).
• The model demonstrated superior AUC compared to 3DMeT across all sequences (P < .001).
• Highest sensitivity (83%) and specificity (83%) were observed for the COR-IW-TSE sequence.

Conclusions:

• MR-Transformer exhibits state-of-the-art performance in predicting knee OA progression to TKR using MRI.
• The model's ability to integrate ImageNet pretraining and 3D spatial information enhances predictive capabilities.
• MR-Transformer represents a significant advancement in AI-driven prognostic tools for knee osteoarthritis.