Artificial intelligence is making big strides in a variety of medical fields, including radiology, oncology, and even ophthalmology. Now a company called Wision AI, based in Shanghai, Cina, is adapting artificial intelligence vision software to help doctors spot polyps during a colonoscopy. The technology is meant for real-time use and the procedure itself doesn’t change much from existing colonoscopies.
We spoke with JingJia Liu, Cofounder and CEO of Wision AI about the company technology, how it works, and what it is capable of already.
Medgadget: Can you briefly describe your company’s technology and how it’s used to detect polyps during colonoscopies? Is it intended to be used before or after a full analysis by a physician?
JingJia Liu: Our software detects colorectal polyps in real-time during the colonoscopy. It acquires a parallel video stream from the endoscope video processor or its monitor, and the algorithm detects the presence of a polyp and displays the findings on a second monitor. It is intended to be used simultaneously by the physician during the procedure analysis.
Medgadget: In your latest press release, you mention that your technology is “built on the same network architecture used to develop self-driving cars, the Wision AI algorithm is designed to enable “self-driving” in colonoscopy procedures.” Can you give provide more insight about how the same technology is being used in such varying fields and if developments in self-driving tech helped your company reach the latest milestone?
JingJia Liu: We used similar deep learning network as self-driving because we have similar real time processing speed requirement as self-driving. Of course, to detect a polyp inside the colon is very different technology from detecting general objects in the street and our technology takes that into account.
Medgadget: Can you briefly describe latest study published in Nature Biomedical Engineering and the results?
JingJia Liu: The algorithm was trained with 5545 colonoscopy images, and validated in three prospectively collected clinical datasets and one open-source dataset. The validation datasets were more than 200 folds in volume than the development dataset. The algorithm achieved > 90% sensitivity and 95% specificity on a per image/frame base in the colonoscopy images and videos. And the system has been proved to be clinically applicable in real time processing during a colonoscopy.
Medgadget: Do you envision that one day colonoscopies will be analyzed primarily by computers?
JingJia Liu: We don’t envision this at the moment. At this stage, our AI for detecting colon polyps is aimed to aid clinicians find more polyps during a routine colonoscopy. In the current clinician-designed clinical trials, our AI technology is only used as a CADx tool for the clinicians operating colonoscopy, rather than a primary analytical robot taking the role of human clinicians.
Medgadget: Have you thought of other medical fields to apply your vision algorithms?
JingJia Liu: In addition to endoscopy, we are applying our technology to other optical medical imaging, such as fundus images and pathological images. But the early cancer screening technology in GI endoscopy is our primary focus. Our next AI product would be the early detection of esophageal squamous cell carcinoma.
Medgadget: What are the next steps for your company?
JingJia Liu: To further validate our AI technology for detecting colon polyps in large-scale and rigorously designed clinical trials, ultimately incorporating this technology into daily practice for improving the quality of colonoscopy in overall colon cancer screening and prevention. Meanwhile, we are actively developing AI tools in other areas and intend to bring them into daily clinical practice improving the quality of current standards there as well.
Medgadget: What impact do you see AI, machine vision, and related technologies having on medicine in the near future?
JingJia Liu: We think the effectiveness of medical/clinical practice can be improved by AI and its related technologies in the near future, but we don’t envision AI to replace the role of human clinicians. From a mathematical point of view, deep learning is bunch of fitting functions with millions of parameters, which certainly cannot “learn” in the same way like human beings.
Being an AI software company, we first need to be truthful about the limitation of this technology, and utilize it to solve problems we believe is can solve.
Link: Wision Ai…