YOLOv8 deep learning model matches senior radiologists in detecting acetabular labral tears on hip MRI

A YOLOv8 deep learning model detected acetabular labral injury on hip MRI with accuracy 0.88, sensitivity 1.00, and specificity 0.74, performing comparably to senior radiologists (P=0.864) and outperforming juniors (P=0.017).

• On a test set of 92 images from 12 patients, YOLOv8m achieved sensitivity 1.00 (95% CI 0.96-1.00) and specificity 0.74 (95% CI 0.62-0.82).
• Model performance was not significantly different from senior chief radiologists (P=0.864) and significantly better than junior radiologists (P=0.017).
• Misjudgment rate on a labrum-absent test set of 58 images was 12.1%.

Automated summary

RadPigeon summaries are original and for information only. They are not clinical advice.