What happened
Wuhan EndoAngel Medical Technology built EA PC-UGI, software that uses deep learning to flag image features of high-grade intraepithelial neoplasia or worse. It works on frozen images of focal gastric lesions taken with magnifying electronic chromoendoscopy. The Center for Medical Device Evaluation (CMDE) of China's National Medical Products Administration reviewed the company's application for Class III registration. The trial was the maker's own, a retrospective, fully crossed multi-reader multi-case study at four sites in China. Its sample size was 340 cases, with pathology as the reference. Readers with the software reached a lesion-level sensitivity of 85.80%, against 79.94% alone, a difference of 5.86% (95% CI 1.72% to 9.99%). Case-level specificity was 80.87% against 78.31%, which met the non-inferiority margin. The AFROC-AUC was 0.865 against 0.831. On 27 January 2026, CMDE concluded that “该产品的上市为适用人群带来的受益大于风险” (putting the product on the market brings the intended population benefits that outweigh the risks). It recommended approval. The readers worked on stored images, so the software guided no care.
Sources
-
Regulatory document, by authors with no tie to the system.
试验组(阅片者在软件辅助下阅片)识别病变的病灶水平的灵敏度及其 95%CI 为 85.80%(82.45%,89.15%)
经综合评价,在目前认知水平上,认为该产品的上市为适用人群带来的受益大于风险。
Details
AI system
- Reported configuration
-
Release V1
- Workflow
-
The software ran only on frozen images that it recognized as taken inside the digestive tract in a magnifying electronic chromoendoscopy mode. It gave a triage prompt on whether the image showed features of high-grade intraepithelial neoplasia or worse.
Follow-up
- Conditions for replication
-
A doctor has judged that a focal gastric lesion needs magnified electronic chromoendoscopy. Images come from the specified endoscopes and image processors. Trained doctors make the diagnosis and do not base a decision on the prompt alone.
Entry details
- Sector
-
Healthcare.
- Purpose
-
Wuhan EndoAngel Medical Technology built EA PC-UGI to flag features of high-grade intraepithelial neoplasia or worse in frozen, magnified endoscopy images of focal gastric lesions.
29 more questions not yet checked
model, provider, role, configuration, inputs and outputs, autonomy, human oversight, beneficiaries, reach, distribution and access, duration, every measurement question, every money question, ownership and control, dependencies, cross-border connections, geopolitical implications, replication attempts, availability, institutional decision, grounds, action and result, success criterion, every evidence question, observed or intended, alternative assessments, related entries, other identifiers, category mappings.
Cite this record
Peter H. Vartanian. (2026). EA PC-UGI helped readers spot more stomach lesions in a trial (AIBD-0958). Artificial Intelligence Benefits Database, edition of 1 November 2026. https://petervartanian.xyz/aibd/records/AIBD-0958.html (archived at https://doi.org/10.7910/DVN/AADSII, version 7.0)