Wision A.I. Announces FDA and CE-MDR Approvals for EndoScreener, Paving the Way for Global Commercialization

In November 2021, EndoScreener (an AI-assisted diagnostic software for digestive endoscopy), the flagship product developed by Wision A.I., successively obtained CE-MDR and FDA approvals, gaining near-simultaneous market access in Europe and the United States.

It is reported that,Wision A.I.’s colonoscopy-assisted diagnostic software is the first Class II registration certificate among similar products since the mandatory implementation of the European Union’s new Medical Device Regulation (MDR), and it is currently the only standalone Software as a Medical Device (SaMD) in this category.

The Medical Device Regulation (MDR), which replaces the previous Medical Devices Directive (MDD) that did not require clinical study evidence, subjects products to more detailed and stringent regulatory scrutiny for certification. Wision A.I.’s Class II CE certificate, issued by the EU-authorised notified body BSI, further underscores its high value and credibility.

Two weeks after obtaining CE MDR certification, Wision A.I. received FDA approval, becoming the first medical AI product approved by the FDA based on clinical evidence from U.S.-based randomized controlled trials.The FDA has previously approved dozens of medical AI applications; however, the vast majority of these approvals were based on validation using retrospective datasets, with only a few supported by prospective single-arm clinical trials or randomized controlled trials conducted outside the United States (OUS). Consequently, the agency has been criticized by the industry for its “lax” regulatory standards.^【1】。

This product was hailed by the FDA as a “Game Changer” during the pre-submission phase. The FDA imposed unprecedentedly high standards for clinical evidence, requiring both crossover and parallel randomized controlled trials. Within two years, MicroVision Medical’s EndoScreener became the first in its class to complete these trials in the United States and meet all endpoint requirements, thereby becoming the first approved standalone software medical device in this category.

Furthermore, unlike medical AI for static images, medical AI-assisted diagnostic systems for real-time dynamic video are considered to pose higher risks and present greater challenges in clinical evaluation from a regulatory perspective.The approval of EndoScreener for market launch by both the EU CE and the US FDA signifies that a methodological framework for real-time dynamic video-based medical AI applications has been established in the field of international medical device regulation.

EndoScreener, an AI-assisted diagnostic software designed for real-time lesion detection and synchronous physician alerts during colonoscopy, has sequentially obtained CE marking and FDA clearance, underpinned by robust clinical research.

Since 2019, EndoScreener has yielded frequent clinical research achievements, setting multiple global firsts.The prospective study published in GUT in 2019 was the first randomized controlled trial (RCT) in the entire field of medical AI; the study published in The Lancet Gastroenterology & Hepatology in 2020 remains the first and only double-blind RCT in the entire field of medical AI.

In September this year, a U.S. multicenter tandem-controlled trial published in CGH became the first independent external randomized controlled trial (RCT) in the global medical AI field. The term “independent external” means that hospitals and physicians who provided training data for the AI were excluded from participating in the RCT to avoid any inherent biases associated with deep learning technologies.

While Wision A.I. created the world’s first, rigorous clinical evidence has also demonstrated from multiple perspectives the significant improvement of EndoScreener on core clinical metrics.Specifically, when physicians used this AI product to assist in colonoscopy, it significantly increased the adenoma detection rate (ADR) and the average number of adenomas detected per procedure (APC), while reducing the miss rate for precancerous lesions.

In the most rigorous and stringent U.S. clinical trials required by the FDA, ten expert physicians from affiliate hospitals of four top-tier medical schools—Harvard University, New York University, the University of Chicago, and Baylor College of Medicine—observed no “ceiling effect” with the use of this product. The integration of EndoScreener still increased the adenoma detection rate (ADR) by 33% among these highly skilled endoscopists (0.9000 vs. 1.1947; p=0.0323) and reduced the miss rate for precancerous lesions by 41% (19.13 vs. 32.52; p=0.0047).

Rigorous clinical trials are a critical step in the commercialization of medical AI products. What technologies underpin their outstanding clinical performance?

Wision A.I.’s answer is few-shot deep learning.It may be hard to imagine that the EndoScreener system was trained on merely over 5,000 endoscopic images, approximately half of which were negative samples. Nevertheless, it was validated against a test set of 1.36 million images in preclinical testing and demonstrated superior performance in subsequent prospective randomized controlled trials.

Dr. Xiao Xiao, Chief Scientist at Wision A.I., believes thatThe amount of data required to solve a clinical problem is determined by the inherent complexity of the problem itself; blindly expanding the training sample size merely amounts to guessing at results. A deep learning model is essentially a function with numerous parameters, so developing AI is the process of constructing this function. Treating deep learning as a “superbrain” and engaging in blind worship has consistently been the greatest impediment to technological progress in this field.

From the perspective of training samples, deep learning with large datasets makes it more difficult to control the construction of function parameters. This is because larger datasets require larger networks for fitting, which easily leads to overfitting on both training and test samples. Another typical phenomenon in medical image recognition is that deep learning models can only identify highly typical lesions, or even only detect targets already identified by physicians at the center of the screen. Such algorithmic models achieve so-called performance only on non-randomized, non-continuous cases or case fragments selected or cropped in a biased manner, holding no clinical application value.

In contrast, few-shot deep learning requires logically defining the function itself based on medical knowledge and constructing a sample space using real clinical cases, thereby achieving robust logical control over recognition performance.The Wision A.I. team published a paper in Nature Biomedical Engineering in 2018, detailing the “few-shot deep learning” and “partial tagging” techniques employed during their development process. To date, peers worldwide have been unable to achieve comparable performance in few-shot training and partial tagging.

Furthermore, few-shot deep learning offers significant advantages in controlling R&D costs. Due to its lower sample size requirements, it can effectively manage the high costs and quality challenges associated with data collection and annotation, compared to deep learning approaches that rely on large datasets.

Setting Aside Complex Data, What Does High-Performance AI-Assisted Colonoscopy Detection Actually Mean?

Colorectal cancer (CRC) is a highly prevalent malignant tumor of the digestive tract. According to statistics from the World Health Organization, colorectal cancer ranks third in global incidence and second in mortality among all malignant tumors.

The progression of gastrointestinal cancers is slow, making early diagnosis and screening critically important: they enable the effective detection and management of precancerous lesions, thereby preventing gastrointestinal cancers and improving patient survival rates. According to relevant papers published in The New England Journal of Medicine, each 1% increase in the adenoma detection rate is associated with a 3% reduction in the risk of interval colorectal cancer.^【2】。

The significance of cancer screening lies not in detecting cancer itself, but in the effective identification and management of precancerous lesions, thereby achieving the goal of cancer prevention. Currently, many cancer screening methods primarily target cancer itself, exhibiting low sensitivity and specificity for precancerous lesions.^【3】, and further confirmation with gold-standard testing is required for definitive diagnosis.

Although colonoscopy is the gold standard for colorectal cancer screening, it still has a relatively high miss rate. A meta-analysis by Academician Li Zhaoshen’s team, published in the journal Clinical Gastroenterology and Hepatology (CGH), based on data from 43 studies across different regions worldwide^【4】, the average miss rate for adenomas reached 26%, and the studies did not include statistics on sessile serrated lesions (SSLs), which are more difficult to identify and have a higher potential for malignant transformation.

Medical AI is the intervention that has contributed most to increasing the detection rate of precancerous lesions over the past two decades, and it can significantly improve this situation.

From a clinical perspective, medical AI can significantly improve the detection rate of precancerous lesions such as adenomas and serrated lesions, and reduce the missed diagnosis rate, by virtue of its high sensitivity and specificity for precancerous lesions.^{【5】【6】【7】【8】【9】}. In the recently published U.S. multicenter clinical trial, EndoScreener demonstrated a per-lesion sensitivity of 99.05%, with an adenoma detection rate (ADR) of 50.44%, which is twice the current U.S. screening benchmark of 25%.

From a health economics perspective, medical AI holds significant value in alleviating hospital burdens, accelerating the training of endoscopists, and reducing healthcare expenditures, particularly for primary care institutions with limited screening capabilities.

Medical AI not only assists physicians in performing efficient colonoscopies but also helps them rapidly develop the ability to differentiate lesions during the process, thereby shortening the training period. Furthermore, the cost of treating late-stage cancer is more than ten times that of early-stage treatment and over 300 times the average cost of treating precancerous lesions. By improving detection rates, medical AI can reduce cancer incidence and the volume of treatments required, leading to a significant reduction in healthcare fiscal expenditures in the long run.

So, how large is this market exactly?

According to statistics from the U.S. Centers for Disease Control and Prevention (CDC), the United States currently performs 20 million colonoscopy screenings annually, covering 60% of the age-eligible population. By 2024, coverage is projected to reach 80% of the age-eligible population, with the number of colonoscopies increasing to 24 million, representing a market worth nearly $40 billion per year.

Domestic market demand is even greater and more urgent.China has a high incidence of gastrointestinal cancers and a large patient population, yet the diagnosis rate remains low. The Expert Consensus on Early Diagnosis and Screening Strategies for Colorectal Tumors in China recommends that colorectal cancer screening target individuals aged 40 to 74 years. Based on the latest census data, approximately 520 million people in China require early cancer screening. Assuming a screening coverage rate of 30%, this would amount to 156 million people.

However, the reality is that China’s capacity and quality in providing digestive endoscopy services lag significantly behind those of developed countries, rendering it currently incapable of conducting population-wide screening for the age-eligible demographic. Taking colonoscopy as an example, the current annual service volume in China is approximately half that of the United States. Given the large population base and accelerating aging trend, increasing the supply of high-quality endoscopic examinations is not only an essential pathway to curbing the incidence of gastrointestinal cancers but also an imperative requirement of the national supply-side structural reform.

EndoScreener is a Software as a Medical Device (SaMD) that can be easily deployed on compatible computer platforms. Distributors can flexibly adjust the system based on the specific clinical environments and payment habits of customers in different regions to accommodate various customized hardware solutions.

Having secured market access in Europe and the United States, Wision A.I. is actively preparing to enter markets in Canada, Australia, Brazil, and Southeast Asia.The company is currently prioritizing overseas markets to achieve the commercial launch of EndoScreener, while steadily advancing its progress in the domestic market.

Liu Jingjia, General Manager of Weishi Medical, introduced:“On one hand, our experience in overseas markets will help drive the expansion of our domestic market; on the other hand, EndoScreener entered the green approval channel for Class III innovative medical devices in China a year ago, becoming the first real-time dynamic imaging AI product reviewed by the National Medical Products Administration (NMPA). Given that our product has vast international markets and strong global competitiveness, we hope to use profits from the international market to subsidize the Chinese market after receiving NMPA approval. This will enable EndoScreener to rapidly enter digestive endoscopy centers at all levels across China with high quality and extremely low costs, supporting the goals of Healthy China 2035.”

AI for digestive endoscopy first gained significant traction in 2017, prompting both international and domestic companies to increase their investments in this field. Recently, Google published research in this area, comparing its performance with that of EndoScreener.^【10】, ByteDance has also filed a patent application with the China National Intellectual Property Administration for intestinal polyp recognition.^【11】. The promising prospects for the development of AI in gastrointestinal endoscopy are once again self-evident.

Notably, unlike most companies in the industry, Wision A.I. has not opted for automated interpretation of lesion characteristics under digestive endoscopy; instead, it has chosen to detect and identify glandular duct variations and their severity in histopathology.

Based on accumulated clinical trial data, Wision A.I. believesThe integration of macroscopic endoscopic imaging findings with microscopic histopathological features will be the pathway to addressing the core clinical challenges in gastrointestinal cancer screening.

Based on accumulating clinical trial evidence regarding AI-assisted colonoscopy, experts in Europe and the United States are already discussing adjustments to the surveillance intervals for patients with positive screening results as specified in current clinical guidelines. A further step involves developing novel colorectal cancer prevention and control models for precision screening and surveillance, tailored to individual patient profiles, examination quality, and histological details. The company has initiated pre-submission discussions with the U.S. Food and Drug Administration (FDA) concerning its innovative digital therapeutics product pipeline.

Therefore, from a broader perspective, Wision A.I. is essentially a digital therapeutics company.Leveraging its strengths in optical imaging AI and complex data modeling, along with its new product technologies and emerging clinical evidence, Wision A.I. is poised to profoundly transform the ways in which people prevent, detect, diagnose, and treat cancer and precancerous lesions.

Wision A.I. was established in October 2016 and currently operates in Shanghai and Chengdu. As an AI company that diverges from mainstream AI technological paradigms, its technical approach is shaped by the integration of cutting-edge mathematical advances in harmonic analysis with contemporary parallel computing architectures. As a healthcare company founded without the participation of senior medical professionals as co-founders, its business philosophy stems from a deep understanding and practical application of rigorous evidence-based medicine and the lifecycle of technology products.

Wision A.I. has achieved a series of medical innovations with a lean organizational structure, while also obtaining regulatory certifications for its medical device quality management system, including China’s YY 0287, the EU’s ISO 13485, and the US MDSAP.

The company’s external investors include Northern Light Venture Capital, Xinruiyuan Capital, and angel investor Bai Xiangbo.

References:

【1】https://khn.org/news/a-reality-check-on-artificial-intelligence-are-health-care-claims-overblown/

【2】Zhao S, Wang S, Pan P, et al. Magnitude, Risk Factors, and Factors Associated With Adenoma Miss Rate of Tandem Colonoscopy: A Systematic Review and Meta-analysis. Gastroenterology. 2019;156(6):1661-1674.e11. doi:10.1053/j.gastro.2019.01.260

【3】 Imperiale TF, Ransohoff DF, Itzkowitz SH, et al. Multitarget stool DNA testing for colorectal-cancer screening. N Engl J Med 2014;370:1287-97.

【4】 Zhao S, Wang S, Pan P, et al. Magnitude, Risk Factors, and Factors Associated With Adenoma Miss Rate of Tandem Colonoscopy: A Systematic Review and Meta-analysis. Gastroenterology. 2019;156(6):1661-1674.e11. doi:10.1053/j.gastro.2019.01.260

【5】Wang P, Berzin TM, Glissen Brown JR, et al. Real-time automatic detection system increases colonoscopic polyp and adenoma detection rates: a prospective randomised controlled study. Gut. 2019;68(10):1813-1819. doi:10.1136/gutjnl-2018-317500

【6】 Liu P, Wang P, Glissen Brown JR, et al. The single-monitor trial: an embedded CADe system increased adenoma detection during colonoscopy: a prospective randomized study. Therap Adv Gastroenterol. 2020;13:1756284820979165. Published 2020 Dec 15. doi:10.1177/1756284820979165

【7】 Wang P, Liu X, Berzin TM, et al. Effect of a deep-learning computer-aided detection system on adenoma detection during colonoscopy (CADe-DB trial): a double-blind randomised study [published correction appears in Lancet Gastroenterol Hepatol. 2020 Apr;5(4):e3]. Lancet Gastroenterol Hepatol. 2020;5(4):343-351. doi:10.1016/S2468-1253(19)30411-X

【8】 Wang P, Liu P, Glissen Brown JR, et al. Lower Adenoma Miss Rate of Computer-Aided Detection-Assisted Colonoscopy vs Routine White-Light Colonoscopy in a Prospective Tandem Study. Gastroenterology. 2020;159(4):1252-1261.e5. doi:10.1053/j.gastro.2020.06.023

 【9】Glissen Brown JR, Mansour NM, Wang P, et al. Deep Learning Computer-aided Polyp Detection Reduces Adenoma Miss Rate: A United States Multi-center Randomized Tandem Colonoscopy Study (CADeT-CS Trial) [published online ahead of print, 2021 Sep 14]. Clin Gastroenterol Hepatol. 2021;S1542-3565(21)00973-3. doi:10.1016/j.cgh.2021.09.009

【10】 Livovsky DM, Veikherman D, Golany T, et al. Detection of elusive polyps using a large-scale artificial intelligence system (with videos). Gastrointest Endosc. 2021;94(6):1099-1109.e10. doi:10.1016/j.gie.2021.06.021

【11】https://www.163.com/dy/article/GO9Q6EDN0519QIKK.html