In the Journals

AI system improves colorectal neoplasm detection

An artificial intelligence-assisted system improved the identification of colorectal neoplasms, according to findings from a study published in Clinical Gastroenterology and Hepatology.

“Computer-aided diagnosis of endoscopic images using artificial intelligence has attracted attention because of its potential for better accuracy and lower interobserver variability,” Shin-ei Kudo, from the Digestive Disease Center, Showa University Northern Yokohama Hospital, Yokohama, Japan, and colleagues wrote. “With [computer-aided] technology, non-expert endoscopists may more easily achieve accuracy levels sufficient to meet the PIVI threshold.”

In the multi-center study, researchers compared the diagnostic performance of the AI system EndoBRAIN with 30 endoscopist trainees and experts. EndoBRAIN analyzed high-quality stained endocytoscopic images only while the endoscopists analyzed white-light microscopy, endocytoscopy with methylene blue staining and endocytoscopy with narrow-band imaging. Investigators used the findings from pathology analysis as the reference standard.

Regarding stained endocytoscopic images, EndoBRAIN identified colon lesions in patients with 96.9% sensitivity, 100% specificity, 98% accuracy, 100% positive-predictive value and 94.6% negative-predictive value. These EndoBRAIN values were significantly greater than those of endoscopy trainees and experts.

When analyzing narrow-band images, EndoBRAIN differentiated between neoplastic and non-neoplastic lesions with 96.9% sensitivity, 94.3% specificity, 96% accuracy, a 96.9% positive predictive value and a 94.3% negative predictive value. While all these values were significantly higher than those of trainees, only the sensitivity and negative predictive values were significantly higher than those of experts. All other values were comparable between EndoBRAIN and experts.

Considering its quick response and 100% reproducibility, EndoBRAIN could be a powerful tool in routine colonoscopy,” Kudo and colleagues wrote. “However, endoscopists must be able to acquire high-quality images to benefit from EndoBRAIN because we evaluated the system’s performance using only high-quality images in this study.”

Based on these results, the Japanese regulatory body Pharmaceuticals and Medical Devices Agency has approved EndoBRAIN for clinical use in Japan. by Erin T. Welsh

Disclosures: Kudo reports receiving lecture fees from Olympus Corp. Please see the study for all other authors’ relevant financial disclosures.

An artificial intelligence-assisted system improved the identification of colorectal neoplasms, according to findings from a study published in Clinical Gastroenterology and Hepatology.

“Computer-aided diagnosis of endoscopic images using artificial intelligence has attracted attention because of its potential for better accuracy and lower interobserver variability,” Shin-ei Kudo, from the Digestive Disease Center, Showa University Northern Yokohama Hospital, Yokohama, Japan, and colleagues wrote. “With [computer-aided] technology, non-expert endoscopists may more easily achieve accuracy levels sufficient to meet the PIVI threshold.”

In the multi-center study, researchers compared the diagnostic performance of the AI system EndoBRAIN with 30 endoscopist trainees and experts. EndoBRAIN analyzed high-quality stained endocytoscopic images only while the endoscopists analyzed white-light microscopy, endocytoscopy with methylene blue staining and endocytoscopy with narrow-band imaging. Investigators used the findings from pathology analysis as the reference standard.

Regarding stained endocytoscopic images, EndoBRAIN identified colon lesions in patients with 96.9% sensitivity, 100% specificity, 98% accuracy, 100% positive-predictive value and 94.6% negative-predictive value. These EndoBRAIN values were significantly greater than those of endoscopy trainees and experts.

When analyzing narrow-band images, EndoBRAIN differentiated between neoplastic and non-neoplastic lesions with 96.9% sensitivity, 94.3% specificity, 96% accuracy, a 96.9% positive predictive value and a 94.3% negative predictive value. While all these values were significantly higher than those of trainees, only the sensitivity and negative predictive values were significantly higher than those of experts. All other values were comparable between EndoBRAIN and experts.

Considering its quick response and 100% reproducibility, EndoBRAIN could be a powerful tool in routine colonoscopy,” Kudo and colleagues wrote. “However, endoscopists must be able to acquire high-quality images to benefit from EndoBRAIN because we evaluated the system’s performance using only high-quality images in this study.”

Based on these results, the Japanese regulatory body Pharmaceuticals and Medical Devices Agency has approved EndoBRAIN for clinical use in Japan. by Erin T. Welsh

Disclosures: Kudo reports receiving lecture fees from Olympus Corp. Please see the study for all other authors’ relevant financial disclosures.