Development and validation of artificial neural networks model for detection of Barrett's neoplasia: a multicenter pragmatic nonrandomized trial (with video).

BACKGROUND AND AIMS: The aim of this study was to develop and externally validate a computer-aided detection (CAD) system for the detection and localization of Barrett's neoplasia and assess its performance compared with that of general endoscopists in a statistically powered multicenter study by using real-time video sequences. METHODS: In phase 1, the hybrid visual geometry group 16-SegNet model was trained by the use of 75,198 images and videos (96 patients) of neoplastic and 1,014,973 images and videos (65 patients) of nonneoplastic Barrett's esophagus. In phase 2, image-based validation was performed on a separate dataset of 107 images (20 patients) of neoplastic and 364 images (14 patients) of nonneoplastic Barrett's esophagus. In phase 3 (video-based external validation) we designed a real-time video-based study with 32 videos (32 patients) of neoplastic and 43 videos (43 patients) of nonneoplastic Barrett's esophagus from 4 European centers to compare the performance of the CAD model with that of 6 nonexpert endoscopists. The primary endpoint was the sensitivity of CAD diagnosis of Barrett's neoplasia. RESULTS: In phase 2, CAD detected Barrett's neoplasia with sensitivity, specificity, and accuracy of 95.3%, 94.5%, and 94.7%, respectively. In phase 3, the CAD system detected Barrett's neoplasia with sensitivity, specificity, negative predictive value, and accuracy of 93.8%, 90.7%, 95.1%, and 92.0%, respectively, compared with the endoscopists' performance of 63.5%, 77.9%, 74.2%, and 71.8%, respectively (P < .05 in all parameters). The CAD system localized neoplastic lesions with accuracy, mean precision, and mean intersection over union of 100%, 0.62, and 0.54, respectively, when compared with at least 1 of the expert markings. The processing speed of the CAD detection and localization were 5 ms/image and 33 ms/image, respectively. CONCLUSION: To our knowledge, this is the first study describing external (multicenter) validation of AI algorithms for the detection of Barrett's neoplasia on real-time endoscopic videos. The CAD system in this study significantly outperformed nonexpert endoscopists on real-time video-based assessment, achieving >90% sensitivity for neoplasia detection. This result needs to be validated during real-time endoscopic assessment.

Development of a real-time endoscopic image diagnosis support system using deep learning technology in colonoscopy.

Gaps in colonoscopy skills among endoscopists, primarily due to experience, have been identified, and solutions are critically needed. Hence, the development of a real-time robust detection system for colorectal neoplasms is considered to significantly reduce the risk of missed lesions during colonoscopy. Here, we develop an artificial intelligence (AI) system that automatically detects early signs of colorectal cancer during colonoscopy; the AI system shows the sensitivity and specificity are 97.3% (95% confidence interval [CI] = 95.9%-98.4%) and 99.0% (95% CI = 98.6%-99.2%), respectively, and the area under the curve is 0.975 (95% CI = 0.964-0.986) in the validation set. Moreover, the sensitivities are 98.0% (95% CI = 96.6%-98.8%) in the polypoid subgroup and 93.7% (95% CI = 87.6%-96.9%) in the non-polypoid subgroup; To accelerate the detection, tensor metrics in the trained model was decomposed, and the system can predict cancerous regions 21.9 ms/image on average. These findings suggest that the system is sufficient to support endoscopists in the high detection against non-polypoid lesions, which are frequently missed by optical colonoscopy. This AI system can alert endoscopists in real-time to avoid missing abnormalities such as non-polypoid polyps during colonoscopy, improving the early detection of this disease.