Learning osteoarthritis imaging biomarkers from bone surface spherical encoding.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Orthopedics

Background:

• Osteoarthritis (OA) is a degenerative joint disease impacting millions globally.
• Accurate diagnosis and prediction of OA progression are crucial for effective patient management.
• Current diagnostic methods may not fully capture the subtle bone shape changes associated with OA.

Purpose of the Study:

• To develop a deep learning model for diagnosing and predicting knee osteoarthritis (OA) using bone shape features.
• To leverage convolutional neural networks (CNNs) for analyzing spherical bone maps derived from MRI.
• To establish bone shape as a novel imaging biomarker for OA detection and prognosis.

Main Methods:

• A CNN model was trained on segmented 3D knee MRI bone data (femur, tibia, patella) converted into spherical maps.
• Over 41,000 merged spherical maps were used to train an OA diagnosis model.
• The trained model was adapted for OA incidence prediction, forecasting OA development up to 8 years in advance.

Main Results:

• The OA diagnosis model achieved an area-under-the-curve (AUC) of 0.905, with 0.815 sensitivity and 0.839 specificity.
• OA incidence models showed AUCs ranging from 0.841 to 0.646 for predictions from 1 to 8 years.
• These results indicate significant performance in both diagnosing current OA and predicting future onset.

Conclusions:

• Bone shape analysis using deep learning is a viable and effective method for OA diagnosis and prediction.
• This study introduces a novel application of AI in musculoskeletal imaging, highlighting bone shape as a predictive biomarker.
• The methodology holds potential for broader applications in identifying other OA biomarkers and improving patient outcomes.