Gastrointestinal Endoscopy Emerges as the New Frontier for Medical Imaging AI: Tencent Miying, Olympus, and Fujifilm Accelerate Strategic Deployments

When it comes to artificial intelligence (AI) in medical imaging, people often think of computer-aided or automated image interpretation in radiology, such as projects for the automatic detection of pulmonary nodules. The field of AI-assisted diagnosis in radiology has attracted substantial investment in research, development, and commercialization from numerous healthcare institutions, multinational corporations, and startups. However, the scope of medical imaging extends far beyond radiological images, encompassing pathology, fundus photography, and endoscopy. In terms of data scale and growing clinical demand, gastrointestinal endoscopic imaging within endoscopy has emerged as the most sought-after sector, captivating countless AI innovators worldwide.

In the second half of 2017, the field of AI related to digestive endoscopy was remarkably vibrant:

In May, Olympus and Fujifilm of Japan announced a collaboration to develop AI-assisted diagnostic products for the stomach and intestines, aiming to improve the efficiency of gastrointestinal lesion detection and reduce missed diagnosis rates. The two companies plan to leverage 300,000 endoscopic images of the stomach and intestines collected from several hospitals in the Tokyo area, which will first be annotated and then used for deep learning training and AI development, with clinical trials scheduled to begin in 2020. As global leaders in digestive endoscopy and each other’s biggest competitors in the industry, their unexpected partnership in the AI field sends a clearly positive signal for AI-enabled digestive endoscopy products and the industry as a whole.

In July, the Japanese National Cancer Center and NEC Corporation announced the development of an artificial intelligence (AI) system capable of automatically diagnosing polyps during colonoscopy. The AI was trained on 140,000 colonoscopic images and tested on 5,000 additional images, achieving an accuracy rate as high as 98%. It was announced that the system would officially enter clinical trials in 2019.

In August, Tencent launched its Miying product, focusing on AI-powered early screening for esophageal cancer, which was deployed at the People’s Hospital of Guangxi Zhuang Autonomous Region. According to data published on the Tencent Miying website, the system achieves a recognition accuracy of 90%, a sensitivity of 87%, and a specificity of 99%. Currently, early screening for esophageal cancer remains highly challenging, with no particularly effective screening methods available. If Tencent can develop an early screening tool suitable for esophageal cancer, it would undoubtedly represent a significant advancement in the medical field and a major innovation capable of reshaping clinical guidelines.

Also in August, West China Hospital and Sichuan Xishi Yigou Company established the West China-Xishi Medical Artificial Intelligence Center and launched an AI-powered digestive endoscopy product capable of accurately identifying polyps, tumors, and varices under gastroscopy, with accuracy rates reaching 92.7%, 93.9%, and 96.8%, respectively.

In October, Sichuan Provincial People's Hospital delivered a keynote address at the World Congress of Gastroenterology (WCOG) on a large-scale prospective clinical trial of AI-assisted diagnosis in digestive endoscopy. The study validated that the real-time monitoring system developed by Shanghai Wuhe Technology achieved a 100% detection rate for precancerous lesions, with frame-by-frame specificity and sensitivity both exceeding 94%, an AUC value as high as 0.991, and a processing speed of 25 frames per second. This achievement received the sole international award presented by the American College of Gastroenterology (ACG), marking the first time that authoritative figures in the clinical gastroenterology community have recognized the clinical value of an AI-assisted diagnostic product. Subsequently, Harvard Medical School also joined the clinical trials of this product.

In December, Japanese media reported that six hospitals, including Yokohama Hospital in Japan, have leveraged artificial intelligence to accurately characterize intestinal polyps within just 0.3 seconds using 500x magnification endoscopy under Narrow Band Imaging (NBI). These six Japanese hospitals have already enrolled in clinical trials of this technology. Previously, determining the malignancy of polyps via biopsy and pathological analysis required up to a week; this advancement undoubtedly optimizes the diagnostic decision-making process for colorectal cancer.

Although the clinical community maintains a highly conservative and cautious stance toward AI-assisted diagnostic technologies and products, this has not hindered gastrointestinal endoscopy from becoming a hotly contested arena in the field of medical imaging AI. Medical device manufacturers, healthcare institutions, and AI startups from China, the United States, and Japan are actively investing in development and clinical trials, with new products being continuously launched. Moreover, products that have begun to win over the highest authorities in the clinical community are already emerging. The author analyzes the following reasons:

I. The Rising Incidence of Gastrointestinal Cancers: According to World Health Organization statistics, gastric cancer, colorectal cancer, and esophageal cancer—the three most common gastrointestinal malignancies—all rank among the top six most prevalent cancers globally. In China, they rank second, fourth, and fifth, respectively, with their combined incidence far exceeding that of lung cancer. These three digestive

The incidence of gastrointestinal cancers in China is on the rise. Gastrointestinal cancers are all mucosal lesions, and the cure rate for early-stage detection is extremely high, making early screening highly significant. Developed countries such as the United States and Japan have national-level screening programs. For example, the United States has significantly reduced the incidence of colorectal cancer over the past decade through colonoscopy screening for the eligible population. Consequently, the U.S. Centers for Disease Control and Prevention (CDC) plans to expand colorectal cancer screening coverage to 80% of the eligible population in the next decade.

In China’s large Grade 3A hospitals, every gastroenterology department or endoscopy center is overcrowded, with waiting times for scheduling gastrointestinal endoscopies often measured in weeks. Leveraging artificial intelligence to assist in the early detection and diagnosis of digestive tract cancers will significantly improve the efficiency of screening and examinations, thereby addressing the challenges of limited and unevenly distributed medical resources.

II. Gastrointestinal endoscopy is a core component of clinical diagnosis and treatment: Compared with radiological image interpretation, gastrointestinal endoscopy serves as the gold standard for screening and diagnosing gastrointestinal tract lesions and is the primary modality for minimally invasive and non-invasive therapies. Artificial intelligence (AI) products developed around gastrointestinal endoscopy will hold greater core clinical value; for instance, automated detection systems for early-stage cancer and precancerous lesions can directly improve early cancer detection rates, thereby effectively reducing cancer mortality. AI can be integrated into every aspect of gastrointestinal endoscopic diagnosis and treatment, demonstrating stronger potential both as an independent medical device in terms of clinical value and commercial monetization.

III. Rapid Technological Advancements in Digestive Endoscopy: As an innovative platform for non-invasive and minimally invasive procedures, digestive endoscopy enables diagnosis and treatment by accessing the human body through natural orifices. An increasing number of surgical interventions are being migrated to the digestive endoscopy platform; even the resection and treatment of tumors in regions adjacent to organs such as the lungs and gallbladder can be performed under endoscopic guidance. Furthermore, digestive endoscopy offers specialized visualization modalities, including special light sources, magnification, and even microscopic observation. These capabilities provide rich tools for pathological research on diseases and create broader opportunities for the application of artificial intelligence.

Of course, the development of artificial intelligence (AI) in digestive endoscopy also faces certain challenges, such as the scarcity of public datasets, the complex morphological features of lesions, high annotation costs, and stringent real-time performance requirements for products. According to current global statistics based on publicly available data, fewer than 30 research institutions, medical facilities, and companies are currently engaged in the development of AI for digestive endoscopy. The author believes that more organizations will join this field, and these difficulties will be overcome one by one by innovators, driven by the substantial clinical and commercial value.