A new study led by researchers at Radboud University Medical Center in the Netherlands has shown that artificial intelligence (AI) can significantly improve the outcome of colonoscopies. With the help of an AI-powered computer, doctors were able to detect 37 percent more adenomas per colonoscopy in the same amount of time as a standard colonoscopy.
Colorectal cancer is the third most common type of cancer worldwide, according to the World Health Organization. While colonoscopy is the gold standard to screen for colorectal cancer, it is estimated that up to 26 percent of adenomas are missed by doctors during standard colonoscopies.
“While adenomas have the potential to develop into malignant colorectal cancer, they are in themselves precursor lesions and not yet malignant,” said lead author Michiel Maas. “However, it is hypothesized that a large proportion of post-colonoscopy colorectal cancers arise from missed lesions.”
The study, published in The Lancet Digital Health, involved 950 patients that were scheduled for a colonoscopy across 10 hospitals in the U.S., Israel, Germany and the Netherlands.
The researchers tested the performance of a computing device developed by the company Magentiq Eye, which funded the study. The AI-powered device monitors the image output of a colonoscopy in real time and places a purple square on areas that look like potential adenomas, drawing the doctor’s attention to examine it.
According to Maas, this computer assistant acts as an “extra set of eyes” during the procedure, reducing the risk of missing lesions.
The device is designed to integrate with existing colonoscopy equipment and runs deep learning algorithms that have been trained using over four million frames of colonoscopy videos, featuring hundreds of adenomas verified by pathologists.
In addition to increasing the rate of adenoma detection, the researchers showed that the computing device was able to reduce the rate at which adenomas were missed from 36% to 19% in patients who underwent both an AI-assisted and a standard colonoscopy in tandem.
When compared to standard colonoscopies, the AI technology did not result in a higher percentage of false positives—which was measured as the percentage of polyps that were removed during the procedure and later found to not be an adenoma by a pathologist. The use of the computer assistant did not significantly increase the time it takes to perform the procedure.
While several computer-aided detection (CAD) systems are already commercially available, including the one used in this study, their use in medical practice is still limited. This is largely due to the fact that reimbursement for AI-assisted medical examinations is still in development in many countries.
“In my opinion, one of the primary challenges in integrating this technology is assessing its cost-effectiveness and establishing suitable reimbursement models,” said Maas.
The researchers plan to further investigate if the computing device can help determine whether a polyp is an adenoma or not. This would allow doctors to leave harmless polyps in place in order to reduce costs and the risk of complications.
The authors also stated that future research should look at long-term outcomes of AI-assisted colonoscopies, including monitoring how many patients develop colorectal cancer after the procedure compared to those undergoing a standard colonoscopy. Maas added that future studies should also explore how CAD systems could influence the training of medical professionals on colonoscopy skills.
