Retrospective diagnostic accuracy study evaluates ChatGPT 5.1's performance in identifying colorectal lesions, suggesting potential clinical use.
Key Points
This study aims to assess the diagnostic accuracy of ChatGPT model 5.1 in distinguishing adenomatous from non-adenomatous colorectal lesions during colonoscopy.
Retrospective analysis of 93 colorectal lesions from colonoscopy
Optical lesion descriptions analyzed by ChatGPT 5.1
Histopathology served as the reference standard for evaluation.
Sensitivity was 97.1%, specificity was 79.3%, overall accuracy was 86.0%
ChatGPT correctly identified 34 adenomas and misclassified one (false negative)
12 non-adenomatous lesions were incorrectly classified as adenomatous (false positives).