In the Journals

Computer-aided colonoscopy accurately identifies small polyps

Diminutive polyps in the distal colon that do not have to be removed can be reliably identified by real-time, computer-aided diagnosis for colonoscopy, according to data published in Annals of Internal Medicine.

“Computer-aided diagnosis for colonoscopy may help endoscopists distinguish neoplastic polyps (adenomas) requiring resection from non-neoplastic polyps not requiring resection, potentially reducing cost,” Yuichi Mori, MD, PhD, from Showa University Northern Yokohama Hospital, Japan, and colleagues wrote.

Mori and colleagues conducted a single-center, open-label, prospective study to examine the feasibility and accuracy of real-time computer-aided diagnosis with endocytoscopes in evaluating diminutive polyps, or those 5 mm or less, among patients (n =791) undergoing colonoscopy. The researchers compared the performance of computer-aided diagnosis with pathologic diagnosis of the resected specimen.

The best-case scenario assumed that polyps lacking either computer-aided diagnosis or pathology were true positive or false positive, and the worst-case scenario assumed they were true negative or false negative.

Data showed that the feasibility of computer-assisted diagnosis was excellent. It had a 100% (475 of 475) success rate for acquiring endocytoscopic images and a 98.1% (457 of 466) success rate for pathologic prediction.

For identifying diminutive rectosigmoid adenomas, computer-aided diagnosis-stained analysis had a negative predictive value of 96.4% (CI, 91.8%-98.8%) in the best-case scenario and 93.7% (CI, 88.3%-97.1%) in the worst-case scenario. Computer-aided diagnosis with narrow-band imaging had a negative predictive value of 96.5% (CI, 92.1%-98.9%) in the best-case scenario and 95.2% (CI, 90.3%-98.0%) in the worst-case scenario.

“Real-time use of the fully automated computer-aided diagnosis system designed for endocytoscopes can meet the clinical threshold required for the diagnose-and-leave strategy for diminutive, non-neoplastic rectosigmoid polyps, which may help improve the cost-effectiveness of colonoscopy” and potentially save $33 million annually, Mori and colleagues concluded.

In an accompanying editorial, Øyvind Holme, MD, PhD, and Lars Aabakken, MD, PhD, both from the University of Oslo, Norway, wrote that there are limitations of computer-aided diagnosis of colonic polyps that need to be addressed before it can be used in clinical practice, including that it was not able to distinguish hyperplastic from adenomatous polyps proximal to the sigmoid colon.

However, the study by Mori and colleagues showcases an important advancement that may lead to reliable identification of distal small colonic polyps, they wrote.

“To err is human, but computer-aided diagnosis may help us reduce the frequency of human errors,” Holme and Aabakken concluded. – by Alaina Tedesco

Disclosure: Aabakken and Holme report no relevant financial disclosures. Mori reports receiving support from Japan Agency for Medical Research and Development, Olympus and The Japan Society for the Promotion of Science, as well as a patent licensed to Showa University and Cybernet Systems. Please see study for all other authors’ relevant financial disclosures.

Diminutive polyps in the distal colon that do not have to be removed can be reliably identified by real-time, computer-aided diagnosis for colonoscopy, according to data published in Annals of Internal Medicine.

“Computer-aided diagnosis for colonoscopy may help endoscopists distinguish neoplastic polyps (adenomas) requiring resection from non-neoplastic polyps not requiring resection, potentially reducing cost,” Yuichi Mori, MD, PhD, from Showa University Northern Yokohama Hospital, Japan, and colleagues wrote.

Mori and colleagues conducted a single-center, open-label, prospective study to examine the feasibility and accuracy of real-time computer-aided diagnosis with endocytoscopes in evaluating diminutive polyps, or those 5 mm or less, among patients (n =791) undergoing colonoscopy. The researchers compared the performance of computer-aided diagnosis with pathologic diagnosis of the resected specimen.

The best-case scenario assumed that polyps lacking either computer-aided diagnosis or pathology were true positive or false positive, and the worst-case scenario assumed they were true negative or false negative.

Data showed that the feasibility of computer-assisted diagnosis was excellent. It had a 100% (475 of 475) success rate for acquiring endocytoscopic images and a 98.1% (457 of 466) success rate for pathologic prediction.

For identifying diminutive rectosigmoid adenomas, computer-aided diagnosis-stained analysis had a negative predictive value of 96.4% (CI, 91.8%-98.8%) in the best-case scenario and 93.7% (CI, 88.3%-97.1%) in the worst-case scenario. Computer-aided diagnosis with narrow-band imaging had a negative predictive value of 96.5% (CI, 92.1%-98.9%) in the best-case scenario and 95.2% (CI, 90.3%-98.0%) in the worst-case scenario.

“Real-time use of the fully automated computer-aided diagnosis system designed for endocytoscopes can meet the clinical threshold required for the diagnose-and-leave strategy for diminutive, non-neoplastic rectosigmoid polyps, which may help improve the cost-effectiveness of colonoscopy” and potentially save $33 million annually, Mori and colleagues concluded.

In an accompanying editorial, Øyvind Holme, MD, PhD, and Lars Aabakken, MD, PhD, both from the University of Oslo, Norway, wrote that there are limitations of computer-aided diagnosis of colonic polyps that need to be addressed before it can be used in clinical practice, including that it was not able to distinguish hyperplastic from adenomatous polyps proximal to the sigmoid colon.

However, the study by Mori and colleagues showcases an important advancement that may lead to reliable identification of distal small colonic polyps, they wrote.

“To err is human, but computer-aided diagnosis may help us reduce the frequency of human errors,” Holme and Aabakken concluded. – by Alaina Tedesco

Disclosure: Aabakken and Holme report no relevant financial disclosures. Mori reports receiving support from Japan Agency for Medical Research and Development, Olympus and The Japan Society for the Promotion of Science, as well as a patent licensed to Showa University and Cybernet Systems. Please see study for all other authors’ relevant financial disclosures.