West China Hospital's Professor Tang Chengwei Unveils China's First Hardware-Integrated AI-Enabled Gastrointestinal Endoscopy Device Receiving Regulatory Approval

The application of artificial intelligence (AI) in medicine has garnered widespread attention, with AI-assisted digestive endoscopic diagnosis emerging as a focal point of industry research. In clinical practice, endoscopic procedures are performed by endoscopists, while diagnoses rely on the physicians’ visual assessment combined with pathological biopsy results—a process that is inherently characterized by continuous experience accumulation and gradual refinement. AI-assisted endoscopic diagnosis holds promise for improving endoscopists’ detection rates of gastrointestinal lesions, thereby effectively reducing misdiagnoses and missed diagnoses.

After several years of development, AI in healthcare has gradually entered a phase of industry consolidation, shifting from initial novelty to practical implementation. Currently, AI healthcare projects with proven real-world applicability command greater market share and influence.

From September 11 to 13, the “2020 Annual Conference of Chinese Digestive Endoscopy,” hosted by the Chinese Medical Association and its Branch of Digestive Endoscopy, was held online. At the conference, Professor Tang Chengwei, Director-Designate of the Gastroenterology Branch of the Chinese Medical Association and from the Department of Gastroenterology at West China Hospital, delivered an inspiring presentation on “Artificial Intelligence Empowering Digestive Endoscopy.” She shared the latest progress in the digestive endoscopy AI project jointly developed by West China Hospital and Xishi Yigou Medical, drawing strong attention from the industry.

(Professor Tang Chengwei, Department of Gastroenterology, West China Hospital; President-Elect of the Chinese Society of Gastroenterology, Chinese Medical Association)

According to the 2015 cancer statistics published online in CA: A Cancer Journal for Clinicians (CA Cancer J Clin), China is a country with a high burden of gastrointestinal cancers. Gastric cancer alone accounts for approximately 679,100 new cases and nearly 500,000 deaths annually, representing 17.7% of all cancer-related deaths.

As the most direct diagnostic equipment for gastrointestinal diseases, digestive endoscopy has become widely available in China. However, the early diagnosis and treatment of gastrointestinal diseases, particularly malignant tumors, remain unsatisfactory. The primary reason is the lengthy training period required for endoscopists; it takes at least ten years to cultivate an experienced physician. This time-intensive process results in a relative scarcity of qualified professionals, thereby constraining the overall standard of diagnosis and treatment for gastrointestinal diseases.

Professor Tang Chengwei pointed out that the field of digestive endoscopy has a high demand and heavy workload for medical staff. The research and development of artificial intelligence technology and equipment in digestive endoscopy are expected to change this situation.

The goal of AI in digestive endoscopy is to enhance diagnostic efficiency and comprehensively improve diagnostic accuracy. By leveraging AI technology, a new endoscopic diagnosis and treatment system will be established, enabling regional medical centers to fulfill their quality control roles and supporting the implementation of China’s tiered diagnosis and treatment policy.

“The quality of gastrointestinal endoscopic diagnosis and treatment in China is uneven, leading many patients to undergo multiple endoscopic procedures, which increases associated risks and wastes medical resources. The implementation of AI technology in gastrointestinal endoscopy enables healthcare providers at high-tier hospitals to deliver more attentive care to patients, allowing physicians to devote greater energy to minimally invasive treatments; for low-tier hospitals, it helps overcome the limitations of insufficient physician training and experience,” stated Professor Tang Chengwei.

At the conference, Professor Tang Chengwei provided insights into the current status and pain points of AI in digestive endoscopy. She pointed out:

I. In the field of AI for digestive endoscopy, there are numerous academic papers but few commercial products, with limited projects actually implemented in clinical practice.

II. Limitations Persist in Endoscopic Clinical Diagnosis. Physicians typically encounter endoscopic lesions of unknown etiology; AI-assisted devices designed solely to detect specific known lesions are difficult to leverage effectively when confronted with a variety of unknown pathologies.

III. The detection accuracy of AI has not yet surpassed that of endoscopists. Protruding lesions are among the main abnormalities observed during endoscopy, which may include polyps, inflammatory lesions, early or advanced-stage cancers, vascular lesions, leiomyomas, and others. Further differentiation of protruding lesions under endoscopy is not difficult for endoscopists; even in cases where distinguishing a small number of protruding lesions is challenging, AI may not necessarily outperform endoscopists.

“The most critical AI assistance endoscopists need is real-time support in detecting lesions during examinations to prevent missed diagnoses, which is particularly beneficial for physicians with limited experience, heavy workloads, or those experiencing fatigue,” pointed out Professor Tang Chengwei.

In March 2017, West China Hospital and Xishi Yigou Medical Technology Co., Ltd. jointly established the West China-Xishi Medical Artificial Intelligence R&D Center. With Professor Tang Chengwei’s team as the core, both parties engaged in in-depth collaboration on the research and development of artificial intelligence products for digestive endoscopy.

Currently, Xishi Yigou Medical's R&D achievements in the field of AI for digestive endoscopy cover the entire gastrointestinal tract, involving more than one hundred diseases across multiple categories, including tumors, polyps, atrophic lesions, ulcerative lesions, erosive lesions, and vascular lesions.

Professor Tang Chengwei introduced that after three years of exploration and research, the center has achieved significant results in the field of AI for digestive endoscopy. Based on the principles of CNNs (Convolutional Neural Networks), an artificial neural network was designed to convert a large number of images of protruding lesions from endoscopies into complex data information, which was then used to train models for an endoscopic CAD (Computer-Aided Diagnosis) system. This effort successfully led to the creation of the first real-time auxiliary diagnostic device for digestive endoscopy imaging (endoscopic AI device).

Following the completion of nearly 500 clinical trial cases, this real-time AI-assisted diagnostic device for digestive endoscopy imaging received approval from the drug regulatory authorities in April 2020, permitting its clinical application as a medical device. It is also the first digestive endoscopy image-assisted diagnostic device approved in China to date.

As the first hardware-integrated AI product for gastroenterology in China, what are its features?

First, this product meets the design standards, with a detection speed that is non-inferior to that of endoscopists, enabling simultaneous identification of protruding lesions. It demonstrates higher sensitivity for protruding lesions ≥5 mm during colonoscopy, reaching 96.81%.

Secondly, endoscopic AI devices have outperformed design expectations in their role as "assistants," providing a valuable foundation for the establishment of industry standards in future endoscopic AI device research and development.

Finally, this product enables physicians to detect lesions in real time with high efficiency. The product series achieves rapid and efficient identification of various lesions throughout the entire gastrointestinal tract (both upper and lower), effectively addressing critical clinical pain points. In terms of disease coverage and clinical commercialization, among other aspects, this product holds a leading international position.

Professor Tang Chengwei introduced that AI-assisted diagnostic devices for digestive endoscopy images are merely an initial product. The application of artificial intelligence in the field of digestive endoscopy is by no means limited to the real-time detection of one or a few types of lesions; instead, broader applications will progressively cover various stages of digestive endoscopic examinations.

Current research findings have matured, and the second-generation product co-developed with Xishi Yigou Medical has largely been finalized. She pointed out that the second-generation product will integrate more closely with clinical practice, penetrating more aspects of digestive endoscopy. On one hand, it achieves new breakthroughs in real-time video recognition of multiple lesions; on the other hand, it introduces more technologies for lesion characterization, quantification, and quality control, thereby realizing a comprehensive layout for both the front-end and back-end of digestive endoscopic examinations.

Since 2017, AI has become a national strategy and played a significant role in the diagnosis of skin cancer and diabetic retinopathy. We believe that AI-based endoscopic image diagnosis and recognition will play an important role in the screening of early gastrointestinal cancers through digestive endoscopy in the future.

Yet, as stated at the outset, after several years of development, the value of medical AI has long been recognized by physicians. However, the concept has long passed its “novelty phase”; only projects that are genuinely applied in clinical scenarios to solve practical problems can command true market influence.