Deep-learning-based clinical decision support system for gastric neoplasms in real-time endoscopy: Development and validation study

Aims his study aimed to establish a deep-learning-based clinical decision support system (CDSS) for the automated detection and classification (diagnosis and invasion-depth prediction) of gastric neoplasms in real-time endoscopy.

Methods A prospective multicenter validation was conducted using 3,976 novel images from five institutions for the classification models. The primary outcomes were the detection rate for the lesion-detection model and accuracy for the lesion-classification model. Clinical benefit was evaluated with 1,210 real-time procedures in a randomized pilot study. Consecutive patients were allocated either to CDSS-assisted screening endoscopy or conventional screening endoscopy. All endoscopic examinations were performed by an expert endoscopist [1] [2] [3].

Results The lesion-detection rate was compared between the groups. The lesion-detection rate was 95.6%, and the mean average precision was 90.6% under the threshold of intersection over union 0.2 in the internal test. The established model reached 81.5% external-test accuracy in the four-class (advanced gastric cancer, early gastric cancer, dysplasias, and non-neoplasm) histopathology prediction. The binary prediction performance (mucosa-confined or submucosa-invaded) of the invasion depth of the detected lesions showed 86.4% external-test accuracy. In real-clinic applications, CDSS-assisted screening endoscopy showed higher lesion-detection rate, but not significant (2.0% vs. 1.5%, P-value = 0.52).

Conclusions We established and tested the deep-learning-based CDSS system for gastric lesions in endoscopic procedures.

 

Scroll to top

Conflicts of interest

Authors do not have any conflict of interest to disclose.

• 1 Cho BJ, Bang CS, Park SW. et al. Automated classification of gastric neoplasms in endoscopic images using a convolutional neural network. Endoscopy 2019; 51: 1121-1129
• 2 Cho BJ, Bang CS.. Artificial Intelligence for the Determination of a Management Strategy for Diminutive Colorectal Polyps: Hype, Hope, or Help. Am J Gastroenterol 2020; 115: 70-72
• 3 Cho BJ, Bang CS, Lee JJ. et al. Prediction of Submucosal Invasion for Gastric Neoplasms in Endoscopic Images Using Deep-Learning. J Clin Med. 2020 9.

Publication History

Article published online:
14 April 2023