The detection of small polyps in real-time to assist GI doctors while performing colonoscopies can improve the diagnosis of these often pre-cancerous lesions. A prospective study showed that AI can be accurate in helping to find small polyps. A team of researchers from Harvard Medical School and Beth Israel Deaconess Medical Center have been working with the Sichuan Academy of Medical Sciences and the Sichuan Provincial People’s Hospital in China, and they’ve been able to use machine learning to spot adenomatous polyps during colonoscopies.
Announced by Chinese AI vendor Shanghai Wision AI Co, the findings could ultimately lead to “self-driving” during colonoscopy procedures. The model used 5,545 images (65.5% of which contained polyps and 34.5% of which didn’t) from the de-identified colonoscopy reports of nearly 1,300 patients. The algorithm was validated using four independent datasets (two for image analysis and two for video analysis). Researchers have said that this deep learning based system has a high performance both with colonoscopy images and real-time videos.
Being able to detect and remove pre-cancerous polyps during a colonoscopy is the ideal way to prevent colon cancer. However, the research shows that the “miss rate” for the more than 14 million colonoscopies carried out each year in the US for adenomas is 6-27%. That’s because it’s hard for clinicians to visually recognize these polyps. That’s where the Wision AI algorithm comes in, acting as a “second set of eyes” and boosting detection rates by as much as 30%.
This is an example of an application of AI that augments (not replaces) the physician performing a procedure. It increases the accuracy and probably the speed of the procedure and happens while AI in monitoring the images in the background. It does not add any extra burden to the performing physician and potentially makes their job easier. For the various AI applications to gain traction in the everyday practice of medicine, it is no enough for AI to lead to better outcomes. It is essential that it would be able to integrate into the clinical workflow and not create extra work for the clinical team.