The doctor slides the colonoscope in and watches the screen like a driver in fog — a pink, shifting tunnel where a dangerous polyp can hide between folds. Something flickers at the edge of the image. A soft box appears, marking a spot the human eye nearly missed.
That marker is not another doctor. It is software, cleared by China's drug regulator, reading the procedure video frame by frame.
A second pair of eyes on the screen
In June 2023, China's National Medical Products Administration (NMPA) approved Tencent Miying (觅影)'s "colon polyp electronic endoscope image detection software" as an innovative medical device. The system takes the live video from the endoscope's image processor, runs a deep-learning model over each frame, and draws a box around areas that look like polyps — in real time, while the doctor is still looking.
Crucially, it does not replace the doctor's view. The original video is unchanged; the marker is an overlay the doctor can accept or ignore. The final call stays with the physician.
What the evidence says
Clinical evaluation was carried out at Zhejiang Provincial People's Hospital and Wenzhou Central Hospital. According to the NMPA announcement and state-media reporting, colonoscopy supported by the system detected more polyps than conventional colonoscopy alone.
That matters because finding polyps early is the whole point of the exam. Most colorectal cancers grow from polyps over years, so removing a polyp during the procedure can prevent a future tumor. China's colorectal cancer burden is large: GLOBOCAN 2020 data cited by Chinese state media put new cases at about 555,000 in 2020, up from 388,000 in 2015.
Why an algorithm helps here
Polyp detection is exhausting and uneven. A doctor's ability to spot a lesion drops with fatigue, experience and the difficulty of the case. Two endoscopists looking at the same video can disagree. An AI model applies the same attention to the first minute and the fortieth, which is exactly where humans slip.
Tencent Miying has built this on years of medical-imaging work; since 2018 it has been part of a national next-generation AI innovation platform for medical imaging. This is a clear case of AI medical imaging (AI医学影像) moving from research papers into a regulated clinical tool.
Where it is used, and what it costs patients
The colon product is deployed in hospitals as part of Tencent Miying's digestive-disease line. The company also holds NMPA registrations for other conditions, including a glaucoma fundus screening system reported to run in over a hundred medical institutions, where it has helped flag previously undiagnosed cases in county hospitals. Those additional deployments rest on the company's own and media reports rather than direct NMPA documentation, so treat them as indicative.
For patients, the AI does not usually add a separate bill — it runs inside the colonoscopy the hospital already performs. A screening colonoscopy at a Chinese public hospital commonly costs around RMB 400 (about US$56 / HK$435), excluding any pathology. The value is in catching more lesions during the one procedure a patient actually shows up for, especially in under-staffed county hospitals where an extra pair of consistent eyes matters most.
What patients should understand
The tool is meant to support, not decide. A marker on the screen is a suggestion the doctor weighs against what they see, the patient's history and the limits of the view. Good deployments pair the software with training so physicians trust it appropriately — neither ignoring the box nor deferring to it blindly. As with any screening, a clean AI-flagged pass is not a lifetime guarantee; guidelines still call for regular repeat exams based on age and risk.
Key takeaways
- A regulator-approved AI now marks suspected colon polyps on the live endoscopy video in Chinese hospitals.
- The tool is an overlay that supports the doctor; the physician still makes the final diagnosis.
- Its biggest benefit is consistency — the same attention across a long, tiring procedure.
What readers can do now
- Ask at booking: if you are over 45 or have family history, ask the hospital whether their endoscopy uses AI detection support — it may improve polyp catch rate.
- Don't skip the prep: the AI is only as good as the view; a poorly cleaned bowel hides polyps from both human and model.
- Keep the report: request the images and findings so a second hospital can review them if anything looked uncertain.
Honest limitations
The headline approval is solid — NMPA's own announcement confirms the colon product. The specific "detection rate improved" claim is reported by state media summarizing the manufacturer's clinical evaluation, not an independent trial we could re-read. Other Tencent Miying approvals (glaucoma, etc.) are mentioned in company and media reports and were not verified against NMPA records here. The AI reduces missed lesions but does not eliminate them, and it depends on a clean viewing field and a doctor who acts on the marker. No tool here should be read as a substitute for regular screening.