Computer-Assisted Diagnosis Brings Gastroenterology into the Future
Despite some skepticism about the usefulness of AI in clinical medicine, there’s ample evidence to show CAD-assisted colonoscopy can save lives.
John Halamka, M.D., president, Mayo Clinic Platform, and Paul Cerrato, senior research analyst and communications specialist, Mayo Clinic Platform, wrote this article.
Gastroenterology is one of the specialties that has shown the most promise in terms of its application of AI and machine learning. For example, Mayo Clinic’s Endoscopy Center is utilizing Mayo Clinic Platform’s resources to explore the value of machine learning in GI care with the assistance of ImaGine, a comprehensive library of endoscopic videos and images linked to clinical data, including symptoms, diagnoses, pathology, and radiology. These data will include unedited full-length videos as well as video summaries of the procedure including landmarks, specific abnormalities, and anatomical identifiers. While these projects have tremendous potential for clinicians “in the trenches” and their patients, even more impressive is the work being done in computer-assisted detection and classification (CADe and CADx).
One of the problems with many of the studies supporting AI-based algorithms is their retrospective design, which is more likely to be flawed because of unanticipated confounding variables. However, in the field of CADe, there have been at least 10 studies that are prospective in nature and all of which were randomized controlled trials, the gold standard in clinical research. The preponderance of evidence indicates that CADe is superior to standard colonoscopy, decreasing adenoma miss rates and increasing adenoma detection rates.
Several companies now make software as a medical device (SaMD) systems that enable better detection rates. The SaMD receives a digital signal from an endoscopy processor and then “outputs a graphical user interface featuring a bounding box at the coordinates of the potential polyp in real time on the existing procedure monitor.” In plain English, that means that as the endoscopist is viewing the patient’s colon on their computer screen, they also see a visual box surrounding the suspicious lesion, enabling the physician to focus their attention on the specific area. The physician can then make a decision about whether the finding is a true or false positive.
Michael Wallace, M.D., from Mayo Clinic’s Division of Gastroenterology and Hepatology, recently spearheaded a clinical research project that demonstrated the value of AI in colonoscopy. He and his colleagues enrolled 230 patients in a randomized trial in which half the group had back-to-back colonoscopies, first with AI algorithms used to assist the diagnosis, followed by the procedure without the help of AI. The other group of patients first had the procedure done without AI, followed by the same procedure with the assistance of AI.
The adenoma miss rate was only 15.5% in the patients who had AI-assisted colonoscopy, compared to 32.4% when the procedure was initially performed without the benefit of AI. Wallace et al. also reported a false negative rate of 6.8% vs 29.6% in the AI and non-AI groups respectively. They concluded: “AI resulted in an approximately 2-fold reduction in miss rate of colorectal neoplasia, supporting AI-benefit in reducing perceptual errors for small and subtle lesions at standard colonoscopy.”
The AI-based system used in the experiment, GI-Genius, relied on a convolutional neural network (CNN) that was previously trained on over 2,600 polyps that were confirmed histologically. In 2021, the Food and Drug Administration approved the CNN to assist clinicians in detecting colonic lesions in real time during a colonoscopy.
As we have pointed out in previous articles, AI cannot replace an experienced physician’s clinical judgement, but studies like this demonstrate that it can have a very real impact when used to augment that judgement.