Computer-Aided Detection AI Reduces Interreader Variability in Grading Hip Abnormalities With MRI.

Area of Science:

• Radiology
• Artificial Intelligence
• Medical Imaging

Background:

• Hip MRI interpretation is challenging, time-consuming, and lacks standardized grading.
• Existing methods suffer from inter- and intra-reader variability.
• A universal grading scale for hip abnormalities is needed.

Purpose of the Study:

• Develop and evaluate a deep learning model for binary classification of hip osteoarthritis (OA) morphological abnormalities on MR images.
• Assess if an AI-based tool improves interreader agreement in hip grading.

Main Methods:

• Retrospective study utilizing 764 hip MRI volumes from 364 patients.
• Trained a deep learning model (MRNet) for classifying cartilage, bone marrow edema-like, and subchondral cyst-like lesions.
• Evaluated interreader agreement before and after AI tool implementation.

Main Results:

• Deep learning model achieved high AUCs for lesion classification: 0.80 (cartilage), 0.84 (edema), 0.77 (cysts).
• AI predictions improved interreader balanced accuracy across all assessed pathologies.
• Interreader agreement increased from 53%, 71%, 56% to 60%, 73%, 68% respectively.

Conclusions:

• Deep learning models demonstrate high performance in classifying hip OA on MR images.
• AI-assisted interpretation significantly improves interreader agreement for hip pathologies.
• This AI tool offers potential to enhance diagnostic accuracy and consistency in hip MRI analysis.