Deep learning detects rotator cuff calcific tendinopathy on shoulder X-ray with AUC ~0.94–0.96 across two centers

AI flags calcific tendinopathy on shoulder radiographs: CNN (VGG19) hit AUC 0.956 internal, 0.940 external; hybrid CNN-ML was similar (0.961/0.942, no significant difference). Trained on 4,268 XRs. Prospective real-world validation still needed.

• Multi-center retrospective study; training set 4,268 balanced shoulder XRs, internal validation n=480, external n=308 — no prospective or reader-comparison data yet.
• End-to-end CNN and hybrid CNN-ML performed equivalently by DeLong test at both sites; CNN favored for workflow simplicity and Grad-CAM saliency interpretability.
• Key limitation: evaluation used balanced datasets, not routine prevalence; no direct comparison with radiologist readers; external prospective validation required before clinical use.

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