Artificial Intelligence Versus Expert Endoscopists for Diagnosis of Gastric Cancer

A Single-center, Retrospective, Open Label, Randomized Controlled Trial of Artificial Intelligence Versus Expert Endoscopists for Diagnosis of Gastric Cancer in Patients Who Underwent Upper Gastrointestinal Endoscopy

Title: A single-center, retrospective randomized controlled trial of artificial intelligence (AI) versus expert endoscopists for diagnosis of gastric cancer in patients who underwent upper gastrointestinal endoscopy. Précis: this single-center, retrospective randomized controlled trial will include 500 outpatients who underwent upper gastrointestinal endoscopy for gastric cancer screening and will compare the diagnostic detection rate for gastric cancer of AI and expert endoscopists. Objectives Primary Objective: to evaluate the diagnostic detection rate for gastric cancer of AI and expert endoscopists. Secondary Objectives: to determine whether AI is not inferior to expert endoscopists in terms of the number of images analyzed for diagnosis of gastric cancer and intersection over union (IOU), and the detection rate of diagnosis of early and advanced gastric cancer. Endpoints Primary Endpoint: diagnosis of gastric cancer. Secondary Endpoints: image based diagnosis of gastric cancer and IOU. Population: in total, 500 males and females aged ≥ 20 years who underwent upper gastrointestinal endoscopy for screening of gastric cancer at a single hospital in Japan. Describe the Intervention: AI-based diagnosis of gastric cancer based on upper gastrointestinal endoscopy images. Study Duration: 3 months.

Prior to Study: Total 500: Screen potential subjects by inclusion and exclusion criteria; obtain endoscopy images. Randomization was performed. Intervention: AI diagnosis was performed for 250 patients using upper gastrointestinal endoscopy images, and Expert endoscopists diagnosis was performed for 250 patients by same methods. Primary analysis: Perform primary analysis of primary and secondary endpoints for 250 patients in each group Cross over diagnosis between AI and expert endoscopists was performed. Perform secondary analysis of agreement of gastric cancer diagnosis per images and IOU between AI and expert endoscopists for 500 patients.

• AI-based diagnosis will be performed based on analysis of endoscopic images (Olympus Optical, Tokyo, Japan). The investigators will use the Single Shot MultiBox Detector (SSD), a deep neural network architecture (https://arxiv.org/abs/1512.02325), and an optimal diagnostic cutoff from a prior report2. The AI system reviewed endoscopy images and reported those in which gastric cancer was detected, together with the coordinates (X, Y) of the lesions.

The expert endoscopists are two physicians with experience of more than 20,000 endoscopies. The expert endoscopists will review the endoscopy images of each patient for 5 min. They will then report endoscopy images in which gastric cancer was detected and manually annotate the lesions in those images.

AI-based diagnosis will be performed based on analysis of endoscopic images (Olympus Optical, Tokyo, Japan). The investigators will use the Single Shot MultiBox Detector (SSD), a deep neural network architecture (https://arxiv.org/abs/1512.02325), and an optimal diagnostic cutoff from a prior report2. The AI system reviewed endoscopy images and reported those in which gastric cancer was detected, together with the coordinates (X, Y) of the lesions.

The expert endoscopists are two physicians with experience of more than 20,000 endoscopies. The expert endoscopists will review the endoscopy images of each patient for 5 min. They will then report endoscopy images in which gastric cancer was detected and manually annotate the lesions in those images.

Inclusion Criteria: Males or females aged ≥ 20 years who underwent upper gastrointestinal endoscopy at Tokyo University Hospital during 2018. Informed optout consent, obtained from each patient before completion of the study. Exclusion Criteria: Patients who underwent gastrectomy. Patients who underwent transnasal upper gastrointestinal endoscopy.

Cancer Genome Atlas Research Network. Comprehensive molecular characterization of gastric adenocarcinoma. Nature. 2014 Sep 11;513(7517):202-9. doi: 10.1038/nature13480. Epub 2014 Jul 23.

Hirasawa T, Aoyama K, Tanimoto T, Ishihara S, Shichijo S, Ozawa T, Ohnishi T, Fujishiro M, Matsuo K, Fujisaki J, Tada T. Application of artificial intelligence using a convolutional neural network for detecting gastric cancer in endoscopic images. Gastric Cancer. 2018 Jul;21(4):653-660. doi: 10.1007/s10120-018-0793-2. Epub 2018 Jan 15.

Niikura R, Aoki T, Shichijo S, Yamada A, Kawahara T, Kato Y, Hirata Y, Hayakawa Y, Suzuki N, Ochi M, Hirasawa T, Tada T, Kawai T, Koike K. Artificial intelligence versus expert endoscopists for diagnosis of gastric cancer in patients who have undergone upper gastrointestinal endoscopy. Endoscopy. 2022 Aug;54(8):780-784. doi: 10.1055/a-1660-6500. Epub 2022 May 4.