Automated Segmentation and Classification of Knee Synovitis Based on MRI Using Deep Learning.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Orthopedics

Background:

• Knee synovitis, including rheumatoid arthritis, gouty arthritis, and pigmented villonodular synovitis, requires accurate diagnosis.
• Accurate segmentation of the suprapatellar capsule (SC) and infrapatellar fat pad (IPFP) is crucial for knee joint assessment.
• Distinguishing between different types of knee synovitis can be challenging for radiologists.

Purpose of the Study:

• To develop a deep learning (DL) model for segmenting the SC and IPFP on sagittal proton density-weighted images.
• To utilize the DL segmentation model to classify three common types of knee synovitis.
• To evaluate the diagnostic performance of the DL model compared to radiologists.

Main Methods:

• A retrospective study of 376 patients with pathologically confirmed knee synovitis.
• Training a semantic segmentation model on manually annotated knee MRI scans.
• Using segmentation results, patient sex, and age for synovitis classification via DL.

Main Results:

• The automated segmentation model achieved high accuracy (0.99) on internal and external test sets.
• The DL model outperformed a senior radiologist in classifying knee synovitis on the internal test set (accuracy 0.86 vs. 0.79).
• On the external test set, the DL model performed comparably to or better than senior and junior radiologists (accuracy 0.79 vs. 0.79/0.73).

Conclusions:

• Deep learning models can accurately segment knee structures like the SC and IPFP.
• DL-based segmentation aids in the accurate classification of knee synovitis.
• These DL models show potential to assist and improve radiologic diagnosis of knee synovitis.