Health

First AI Device for Colonoscopy: Extra Set of Expert ‘Eyes’

The first artificial intelligence (AI) endoscopy module developed specifically to help detect adenomas during routine colonoscopy is making its debut following approval by the US Food and Drug Administration (FDA) of the system earlier this month.

The GI Genius module is the first and only commercially available computer-aided detection system that uses AI to identify colorectal polyps during routine colonoscopy.

The technology is compatible with most standard video endoscopy systems and has been “trained” to identify colonic lesions that are possibly cancerous, according to Medtronic, the distributor of the device.

“I think that anything we can do within a reasonable cost that enhances quality and patient outcomes during colonoscopy warrants very close consideration,” David Johnson, MD, professor of medicine and chief of gastroenterology, Eastern Virginia Medical School, Norfolk, Virginia, told Medscape Medical News.

He was not involved with the development of the GI Genius system but has worked with a similar AI device that is used in conjunction with colonoscopy.

“The whole development of the technology for AI is done by inputting repetitive images into the computer, where it develops what is called the ‘neural network,’ ” he explained.

The computer then draws upon the “education” of this neural network to identify different types of colonic lesions — “and the more inputs that are put into the computer to enhance the neural network, the more capable the program becomes in the identification of variants and lesion size and characteristics,” Johnson added.

During routine colonoscopy, the GI Genius system generates visual markers ― essentially, small green squares — and a low-volume sound whenever the software detects a region of interest.

These squares are superimposed on the video generated by the endoscope camera to alert the colonoscopist to regions that may require closer assessment, either visually, by tissue sampling, or by removal of the lesion itself.

“Colonoscopy is a durable screening and surveillance strategy, but it’s not perfect [because] it depends on a physician’s skill and their ability to pick up polyps in the colon,” Jeremy Glissen Brown, MD, fellow, Beth Israel Deaconess Medical Center, Boston, Massachusetts, told Medscape Medical News. He has also worked with an AI device.

Studies of adenoma detection during “all-comer” colonoscopies show that the rate of missed lesions ranges from a low of 6% to 40%, “so polyps are still missed during colonoscopy, and any technology that can solve parts of that problem is welcome,” Glissen Brown commented.

Clinical Trial Data That Led to Approval

The recent FDA approval of the GI Genius device was based on a prospective, randomized trial that was published in Gastroenterology in 2020. That trial involved 700 patients who were being screened or followed with colonoscopy every 3 years of longer. Participants underwent either white-light standard colonoscopy with the assistance of the GI Genius technology or standard white-light colonoscopy alone.

Results showed that the combination of standard colonoscopy and the GI Genius module identified laboratory-confirmed adenomas or carcinomas in 54.8% patients, compared to 40.4% of patients who underwent colonoscopy alone — a difference of 14%.

In the Gastroeneterology article, the authors write that the “14% absolute increase in adenoma detection rate (ADR) obtained by computer-aided detection (CADe) in our study indicates that failure in polyp recognition is a clinically relevant cause of miss rate. Of note, the efficacy of CADe in reversing such miss rate also indicates that the same operator who missed the lesion in the first place was able to correctly diagnose it when the lesion was presented by the CADe. This underlines that the main cognitive challenge in polyp recognition is the discrimination between the candidate lesion and the surrounding healthy mucosa, whereas its correct characterization as neoplastic tissue that occurs after CADe detection is apparently a much easier task.”

The authors also note that they did “not assess the actual number of false positive activations by the system, as this would have altered the routine setting of our study,” but they refer to a study published in Gut in 2020 in which false positive frames were seen in fewer than 1% of frames from the whole colonoscopy.

Because the new device improves on the ability of colonoscopy to detect lesions overall, it can reduce the risk for the occurrence of interval cancers between colonoscopies, Medtronic suggests.

Previous research has shown that every 1% increase in the adenoma detection rate results in a 3% decrease in the risk for colorectal cancer.

“More than 19 million screening colonoscopies are performed in the United State each year…. Detection of adenomas during colonoscopy is an important quality metric,” James Weber, MD, a gastroenterologist affiliated with Texas Digestive Disease Consultants, Southlake, Texas, commented in a Medtronic press release.

“The addition of AI can increase the quality of colonoscopies, potentially improving diagnosis and outcomes for colon cancer patients,” he added.

Weber is also the CEO of GI Alliance, a physician-led national healthcare platform of independent GI practices in six states in the United States.

Computer-Aided Detection

Unlike other computer-aided detection technologies, GI Genius does not characterize or “diagnose” a lesion, nor does it replace laboratory sampling as a means of confirming a cancer diagnosis.

The technology acts essentially as an extra set of expert “eyes” to detect suspicious lesions during colonoscopy, which should prove helpful, Johnson and Glissen Brown both commented.

“When a gastroenterologist looks at the video image, typically, our eyes are focused in the center of that image — that’s where our 20/20 vision is,” Johnson explained.

The computer has 20/20 vision over the whole image, including the periphery, “so the technology really gives an extremely expanded acuity of vision and highlights areas that we may need to investigate further,” he added.

Glissen Brown was involved in a trial of another AI device ― the real-time automatic polyp detection system (Shanghai Wision AI Co, Ltd). That study showed an increase in colonoscopic polyp and adenoma detection rates, but this was mainly due to a higher number of diminutive adenomas detected by the automatic detection system, Gliseen Brown said. There was no important difference in the number of larger adenomas detected with the device and the number detected without it.

However, there was a significant increase in the detection of hyperplastic polyps when the automatic detection system was used. “We definitely want to look at the false positive rate — both the false positive rate under the camera when we are doing colonoscopy and under the microscope when we do biopsies,” Glissen Brown acknowledged.

In numerous prospective studies of various computer-aided detection technologies such as the GI Genius system, the false positive rate resulting in the performance of biopsy of insignificant lesions is relatively low, he said.

“Ultimately, the decision to remove or biopsy a lesion is with the physician, because the GI Genius technology just points the provider to the area of concern, and then it’s up to them to look at it and decide whether it needs to be biopsied or not,” Glissen Brown said.

“So the technology serves more as a digital safety net and points the physician in the right direction, so it shouldn’t lead to much in the way of histologic false positives,” he noted.

The only potential disadvantage to using an AI system such as the GI Genius module is the time it might take for endoscopists to learn how to use it and how much the technology might increase the time required to perform the procedure, he added.

For about 18 months, Johnson has been running a clinical trial with a similar type of AI technology during colonoscopy. He has found that the learning curve for using these systems is “inordinately short.” Glissen Brown agrees and suggests that if physicians are already performing colonoscopies regularly, they could probably learn to use an AI system such as GI Genius in about a week.

In his experience, Johnson has found that the delay caused by use of an AI system during colonoscopy is “minimal.”

If there is any delay at all, “we know that time in the colon on withdrawal increases the detection of polyps, so more time during withdrawal may be a good thing,” he added. It should be noted that endoscopy societies recommend a withdrawal time of at least 6 minutes, which is one of the metrics used to ensure the quality of a colonoscopy, Glissen Brown explained.

Indeed, the pivotal study upon which the FDA approved the GI Genius module required a minimum withdrawal time of 6 minutes. Participants reported that they did not find that using the GI Genius increased withdrawal time, he added.

“I think there is enough prospective evidence at this point to suggest that this technology may really be of benefit to clinicians with a lot of different skill levels, so I would be eager to know how clinicians interact with it in the clinical setting,” Glissen Brown commented.

Johnson agreed, noting, “even the good can get better.”

Johnson has served as a director, officer, partner, employee, advisor, consultant, or trustee for WebMD/Medscape, CRH Medical, the American College of Gastroenterology Research Institute, and HyGIeaCare. Glissen Brown has disclosed no relevant financial relationships.

For more from Medscape Oncology, join us on Twitter and Facebook.


Most Related Links :
Business News Governmental News Finance News

Source link

Back to top button