A multicentre team of researcher-clinicians led by gastroenterologists at Beth Israel Deaconess Medical Center (BIDMC) assessed whether artificial intelligence-based computer aided adenoma detection can improve colonoscopy quality by reducing the miss rate. The researchers reported a relative reduction of the miss rate by nearly a third when computer-aided detection was used in conjunction with standard-of-care colonoscopy.
The study is the first randomised trial examining the role of a deep-learning based computer-aided detection system during colonoscopy in the US and is also one of the first randomised trials examining the role of an artificial intelligence intervention in any field of medicine. The findings were featured in the paper, ‘Deep Learning Computer-Aided Polyp Detection Reduces Adenoma Miss Rate: A U.S. Multi-Center Randomized Tandem Colonoscopy Study (CADeT-CS Trial)’ published in the journal Clinical Gastroenterology and Hepatology.
"Our study demonstrates that computer-aided polyp detection has the potential to decrease variability in colonoscopy quality among providers by reducing the miss rate even for experienced physicians," said senior author, Dr Tyler M Berzin, director of the Advanced Endoscopy Fellowship at BIDMC. "These results suggest that artificial intelligence may be an important tool to help reduce the incidence of colorectal cancer in the US through improvements in screening quality."
For their prospective, multicentre trial, Berzin, also an associate professor of medicine at Harvard Medical School, and colleagues enrolled 223 patients presenting for colorectal cancer screening or surveillance at four academic medical centres in the US from 2019 through 2020. All patients underwent both the standard high-definition, white light colonoscopy and a computer-aided detection assisted colonoscopy.
Half were randomised to undergo the standard colonoscopy first, followed immediately by the other procedure in tandem fashion by the same endoscopist. The other half were randomised to receive the procedures in reverse order.
The team found that for the group that underwent computer-aided colonoscopy first, the adenoma miss rate was just above 20 percent, significantly lower than the 34 percent miss rate among those who received standard high-definition white light colonoscopy first.
"This trial has implications beyond colorectal screenings, as we believe it to be one of the first U.S. prospective randomized control trials evaluating an artificial intelligence technology in any field of medicine," said first author, Dr Jeremy R Glissen Brown, a fellow in the Division of Gastroenterology & Hepatology at BIDMC. "While deep learning for the purposes of artificial intelligence has been area of intense commercial and research focus for the past five years, it is crucial that we collect high-quality, rigorous data in the form of randomized clinical trials in order for us to incorporate AI safely and effectively in medical practice."
To access a journal pre-proof, please click here