ChatGPT-5 Shows Low Sensitivity for Dental Pathology on X-rays Despite High Specificity

ChatGPT-5 had high specificity (99.3%) but poor sensitivity for dental pathology on X-rays: short obturation 13.7%, long 0%, voids 9-23%, periapical lesions 11-29%. Accuracy (54-63%) was much lower than dentists (76-86%, p<0.001).

• Retrospective analysis of 271 periapical radiographs of endodontically treated posterior teeth, comparing ChatGPT-5 to general dentists and an endodontic reference standard.
• Sensitivity for detecting short obturation (13.7%), long obturation (0%), voids (9.0-22.7%), and periapical lesions (10.5-28.6%) was very low, despite high specificity for normal findings (up to 99.3%).
• Limitations: single AI model, no external or prospective validation, generalizability restricted to posterior teeth.

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