YOLOv8 algorithm-aided detection of patellar instability or dislocation on knee joint MRI images

Area of Science:

• Orthopedics
• Medical Imaging
• Artificial Intelligence

Background:

• Diagnosing patellar instability or dislocation is challenging using traditional methods.
• Current imaging techniques like X-ray, CT, and MRI have limitations.
• Computer vision has not been widely applied to patellar instability diagnosis.

Purpose of the Study:

• To evaluate the feasibility of the You Only Look Once (YOLO) algorithm for detecting patellar instability or dislocation.
• To compare the diagnostic performance of YOLOv8 with junior radiologists.

Main Methods:

• A dataset of 550 knee MRI scans was used, with 190 diagnosed with patellar instability or dislocation.
• Four indicators on transverse knee MRI scans were used to identify patellar instability.
• The YOLOv8 algorithm was trained on 450 labeled images and validated on 100 unlabeled images.

Main Results:

• The YOLOv8 model achieved 62% sensitivity, 97% specificity, and 83% accuracy.
• The YOLOv8 model's performance was comparable to a junior radiologist (62% sensitivity, 82% specificity, 74% accuracy).
• YOLOv8 interpreted images in approximately 14 ms, significantly faster than a radiologist (9.55 s).

Conclusions:

• The refined YOLOv8 model is a viable tool for identifying patellar instability or dislocation.
• YOLOv8 demonstrates comparable diagnostic accuracy to junior radiologists.
• The AI model offers a substantial reduction in image interpretation time.