Evaluation of a Computer-aided Diagnosis System (CADx) in the Early Detection of Gastric Cancer in France. Cancer in France.

Overview

Upper gastrointestinal (GI) cancers are one of the most common cancers worldwide. Except for cardia cancers, the incidence of gastric cancer has decreased consistently since 1980, but remains at a high level. In France, gastric cancers are the 6th most common cause of cancer-related mortality. The risk factors of upper GI cancers are well known and their control could prevent the development of cancers: smoking cessation, reduction of obesity, alcohol, eradication of Helicobacter pylori. But late presentation with upper GI cancer results in a poorer prognosis. Patients with advanced (Stage IV) gastric cancer have a five-year survival rate of 3.7% whereas patients whose gastric cancer is discovered in its early stage (Stage I) have a significantly higher five-year survival rate of 88.4%. Therefore, endoscopic detection of upper GI lesions at an earlier stage is the single most effective measure for reducing cancer mortality. But upper GI cancer is also often missed during examinations, and some studies demonstrated a missed cancer rate of 2.3-13.9% in Western populations. In the past decade, accurate diagnosis during endoscopy has become particularly important as dysplastic lesions and early gastric cancers can be treated effectively with both endoscopic mucosal resection (EMR) and endoscopic submucosal dissection (ESD), avoiding the morbidity and mortality associated with gastrectomy. However, these early neoplastic lesions can be sometimes difficult to distinguish from background mucosa, even with advanced imaging techniques (high definition, chromoendoscopy). In recent years, image recognition using artificial intelligence (AI) with deep learning has dramatically improved and opened the door to more detailed image analysis and real time application in various medical field, including endoscopy. For example, in the colorectal cancer screening area, real time computer-aided detection systems (CADe) can lead to significant increases in both polyp and adenoma detection rates. CADe has also shown good performance in detection of Barrett's neoplasia during live endoscopic procedures in order to more accurately locate the area to be biopsied. Recently, a Chinese study showed that CADe achieved high diagnostic accuracy in detecting upper GI cancers, with sensitivity similar to that of expert endoscopists and superior to that of non-experts. This system could support non-experts by improving their diagnostic accuracy to a level similar to that of experts and provide assistance for improving the effectiveness of upper GI cancer diagnosis and screening. Although encouraging results have been published regarding the use of AI in the diagnosis of upper GI cancers, the clinical applicability of such systems in a European population has yet to be investigated. Therefore, we want to evaluate the diagnostic capability of a recent CADx compared to endoscopists in order to improve the real-time detection of early gastric cancers in our European center Edouard Herriot Hospital, Lyon, France, as well as 3 other tertiary centers in France (Limoges, Rennes and Nancy University Hospitals). With a high prevalence of stomach cancer, Japan is a world leader in high-quality diagnostic upper GI endoscopy, and the clinical routine in this country differs substantially from Western practice, with population-based screening programs. We will use for our study a CADx developed by AI medical service Inc. (1-18-1, Higashiikebukuro, Toshima-ku, Tokyo 170-0013, Japan), a Japanese company developing AI systems that supports endoscopist's diagnosis for the digestive tract. A recent study involving AI medical service system showed good results in the diagnosis of early gastric cancer compared to endoscopists, with a significantly higher sensitivity.