Real-time, computer-aided detection system significantly improved adenoma detection

A real-time, computer-aided detection system using artificial intelligence significantly improved adenoma detection during high-definition colonoscopy in a multicenter, randomized clinical trial.

The adenoma detection rate was 55% in the intervention group and 40% in the control group, Alessandro Repici, MD, PhD, and his associates wrote in Gastroenterology. Improved detection of smaller adenomas explained the difference. After age, sex, and indication for colonoscopy were controlled for, computer-aided detection (CADe) increased the probability of adenoma detection by 30% (risk ratio, 1.30; 95% confidence interval, 1.14-1.45).

The CADe system did not increase the likelihood of resecting non-neoplastic lesions (26% versus 29% in the control group), said Dr. Repici, of Humanitas Research Hospital in Milano, Italy. “The per-protocol analysis produced similar results,” he and his associates wrote. “The substantial improvement for adenoma detection rate and mean number of adenomas per colonoscopy, without increasing the removal of nonneoplastic lesions, is likely to improve the quality of colonoscopy without affecting its efficiency.”

Screening colonoscopies miss about 25% of adenomas, increasing patients’ risk for colorectal cancer. Although real-time CADe systems can identify colorectal neoplasias, comprehensive studies of the effect of CADe systems on adenoma detection and other colonoscopy quality measures are lacking.

The study included 685 adults from three centers in Italy who underwent screening colonoscopies for colorectal cancer, postpolypectomy surveillance, or workup based on a positive fecal immunochemical test or signs and symptoms of colorectal cancer. Patients were randomly assigned on a one-to-one basis to receive high-definition colonoscopies with or without the CADe system, which consists of an artificial intelligence–based medical device (GI Genius, Medtronic) that processes colonoscopy images in real time and superimposes a green box over suspected lesions. Six experienced endoscopists performed the colonoscopies; the minimum withdrawal time was 6 minutes, and histopathology was the reference standard.

The average number of adenomas detected per colonoscopy was 1.1 (standard deviation, 0.5) in the CADe group and 0.7 (SD, 1.2) in the control group, for an incidence rate ratio of 1.46 (95% CI, 1.15-1.86). The CADe system also significantly improved the detection of adenomas measuring 5 mm or less (34% vs. 27% in the control group; RR, 1.26; 95% CI, 1.01-1.52) and adenomas measuring 6-9 mm (11% vs. 6%, respectively; RR, 1.78; 95% CI, 1.09-2.86). Detection of larger adenomas did not significantly differ between groups. These findings did not vary based on adenoma morphology (polypoid or nonpolypoid) or location (proximal or distal colon), the researchers said.

Detection of multiple adenomas also was higher in the intervention group than in the control group (23% vs. 15%, respectively; RR, 1.50; 95% CI, 1.19-1.95). There were no significant differences in the detection of sessile serrated lesions (7% and 5%) and nonneoplastic lesions (20% and 17%). Average withdrawal times did not significantly differ between groups (417 seconds for CADe and 435 seconds for the control group).

The CADe system is a convolutional neural network that was trained and validated using a series of more than 2,600 histologically confirmed polyps from 840 participants in a prior clinical trial (Gastroenterology 2019;156:2198-207.e1). The system takes an average of 1.5 microseconds to output processed images.

“The addition of real-time CADe to colonoscopy resulted in a 30% and 46% relative increase in adenoma detection rate and the average number of adenomas detected per colonoscopy, demonstrating its efficacy in improving the detection of colorectal neoplasia at screening and diagnostic colonoscopy,” the investigators wrote. “[The s]afety of CADe was demonstrated by the lack of increase of both useless resections and withdrawal time, as well as by the exclusion of any underskilling in the study period.”

Medtronic loaned the equipment for the study. Dr. Repici and the senior author disclosed consulting fees from Medtronic.

SOURCE: Repici A et al. Gastroenterology. 2020 May 3. doi: 10.1053/j.gastro.2020.04.062.

A real-time, computer-aided detection system using artificial intelligence significantly improved adenoma detection during high-definition colonoscopy in a multicenter, randomized clinical trial.

The adenoma detection rate was 55% in the intervention group and 40% in the control group, Alessandro Repici, MD, PhD, and his associates wrote in Gastroenterology. Improved detection of smaller adenomas explained the difference. After age, sex, and indication for colonoscopy were controlled for, computer-aided detection (CADe) increased the probability of adenoma detection by 30% (risk ratio, 1.30; 95% confidence interval, 1.14-1.45).

The CADe system did not increase the likelihood of resecting non-neoplastic lesions (26% versus 29% in the control group), said Dr. Repici, of Humanitas Research Hospital in Milano, Italy. “The per-protocol analysis produced similar results,” he and his associates wrote. “The substantial improvement for adenoma detection rate and mean number of adenomas per colonoscopy, without increasing the removal of nonneoplastic lesions, is likely to improve the quality of colonoscopy without affecting its efficiency.”

Screening colonoscopies miss about 25% of adenomas, increasing patients’ risk for colorectal cancer. Although real-time CADe systems can identify colorectal neoplasias, comprehensive studies of the effect of CADe systems on adenoma detection and other colonoscopy quality measures are lacking.

The study included 685 adults from three centers in Italy who underwent screening colonoscopies for colorectal cancer, postpolypectomy surveillance, or workup based on a positive fecal immunochemical test or signs and symptoms of colorectal cancer. Patients were randomly assigned on a one-to-one basis to receive high-definition colonoscopies with or without the CADe system, which consists of an artificial intelligence–based medical device (GI Genius, Medtronic) that processes colonoscopy images in real time and superimposes a green box over suspected lesions. Six experienced endoscopists performed the colonoscopies; the minimum withdrawal time was 6 minutes, and histopathology was the reference standard.

The average number of adenomas detected per colonoscopy was 1.1 (standard deviation, 0.5) in the CADe group and 0.7 (SD, 1.2) in the control group, for an incidence rate ratio of 1.46 (95% CI, 1.15-1.86). The CADe system also significantly improved the detection of adenomas measuring 5 mm or less (34% vs. 27% in the control group; RR, 1.26; 95% CI, 1.01-1.52) and adenomas measuring 6-9 mm (11% vs. 6%, respectively; RR, 1.78; 95% CI, 1.09-2.86). Detection of larger adenomas did not significantly differ between groups. These findings did not vary based on adenoma morphology (polypoid or nonpolypoid) or location (proximal or distal colon), the researchers said.

Detection of multiple adenomas also was higher in the intervention group than in the control group (23% vs. 15%, respectively; RR, 1.50; 95% CI, 1.19-1.95). There were no significant differences in the detection of sessile serrated lesions (7% and 5%) and nonneoplastic lesions (20% and 17%). Average withdrawal times did not significantly differ between groups (417 seconds for CADe and 435 seconds for the control group).

The CADe system is a convolutional neural network that was trained and validated using a series of more than 2,600 histologically confirmed polyps from 840 participants in a prior clinical trial (Gastroenterology 2019;156:2198-207.e1). The system takes an average of 1.5 microseconds to output processed images.

“The addition of real-time CADe to colonoscopy resulted in a 30% and 46% relative increase in adenoma detection rate and the average number of adenomas detected per colonoscopy, demonstrating its efficacy in improving the detection of colorectal neoplasia at screening and diagnostic colonoscopy,” the investigators wrote. “[The s]afety of CADe was demonstrated by the lack of increase of both useless resections and withdrawal time, as well as by the exclusion of any underskilling in the study period.”

Medtronic loaned the equipment for the study. Dr. Repici and the senior author disclosed consulting fees from Medtronic.

SOURCE: Repici A et al. Gastroenterology. 2020 May 3. doi: 10.1053/j.gastro.2020.04.062.

A real-time, computer-aided detection system using artificial intelligence significantly improved adenoma detection during high-definition colonoscopy in a multicenter, randomized clinical trial.

The adenoma detection rate was 55% in the intervention group and 40% in the control group, Alessandro Repici, MD, PhD, and his associates wrote in Gastroenterology. Improved detection of smaller adenomas explained the difference. After age, sex, and indication for colonoscopy were controlled for, computer-aided detection (CADe) increased the probability of adenoma detection by 30% (risk ratio, 1.30; 95% confidence interval, 1.14-1.45).

The CADe system did not increase the likelihood of resecting non-neoplastic lesions (26% versus 29% in the control group), said Dr. Repici, of Humanitas Research Hospital in Milano, Italy. “The per-protocol analysis produced similar results,” he and his associates wrote. “The substantial improvement for adenoma detection rate and mean number of adenomas per colonoscopy, without increasing the removal of nonneoplastic lesions, is likely to improve the quality of colonoscopy without affecting its efficiency.”

Screening colonoscopies miss about 25% of adenomas, increasing patients’ risk for colorectal cancer. Although real-time CADe systems can identify colorectal neoplasias, comprehensive studies of the effect of CADe systems on adenoma detection and other colonoscopy quality measures are lacking.

The study included 685 adults from three centers in Italy who underwent screening colonoscopies for colorectal cancer, postpolypectomy surveillance, or workup based on a positive fecal immunochemical test or signs and symptoms of colorectal cancer. Patients were randomly assigned on a one-to-one basis to receive high-definition colonoscopies with or without the CADe system, which consists of an artificial intelligence–based medical device (GI Genius, Medtronic) that processes colonoscopy images in real time and superimposes a green box over suspected lesions. Six experienced endoscopists performed the colonoscopies; the minimum withdrawal time was 6 minutes, and histopathology was the reference standard.

The average number of adenomas detected per colonoscopy was 1.1 (standard deviation, 0.5) in the CADe group and 0.7 (SD, 1.2) in the control group, for an incidence rate ratio of 1.46 (95% CI, 1.15-1.86). The CADe system also significantly improved the detection of adenomas measuring 5 mm or less (34% vs. 27% in the control group; RR, 1.26; 95% CI, 1.01-1.52) and adenomas measuring 6-9 mm (11% vs. 6%, respectively; RR, 1.78; 95% CI, 1.09-2.86). Detection of larger adenomas did not significantly differ between groups. These findings did not vary based on adenoma morphology (polypoid or nonpolypoid) or location (proximal or distal colon), the researchers said.

Detection of multiple adenomas also was higher in the intervention group than in the control group (23% vs. 15%, respectively; RR, 1.50; 95% CI, 1.19-1.95). There were no significant differences in the detection of sessile serrated lesions (7% and 5%) and nonneoplastic lesions (20% and 17%). Average withdrawal times did not significantly differ between groups (417 seconds for CADe and 435 seconds for the control group).

The CADe system is a convolutional neural network that was trained and validated using a series of more than 2,600 histologically confirmed polyps from 840 participants in a prior clinical trial (Gastroenterology 2019;156:2198-207.e1). The system takes an average of 1.5 microseconds to output processed images.

“The addition of real-time CADe to colonoscopy resulted in a 30% and 46% relative increase in adenoma detection rate and the average number of adenomas detected per colonoscopy, demonstrating its efficacy in improving the detection of colorectal neoplasia at screening and diagnostic colonoscopy,” the investigators wrote. “[The s]afety of CADe was demonstrated by the lack of increase of both useless resections and withdrawal time, as well as by the exclusion of any underskilling in the study period.”

Medtronic loaned the equipment for the study. Dr. Repici and the senior author disclosed consulting fees from Medtronic.

SOURCE: Repici A et al. Gastroenterology. 2020 May 3. doi: 10.1053/j.gastro.2020.04.062.