FRIDAY, March 5, 2021 (HealthDay News) — Machine learning-based image analysis facilitates noninvasive differentiation of benign and premalignant colorectal polyps with computed tomography (CT) colonography, according to a study published online Feb. 23 in Radiology.

Sergio Grosu, M.D., from University Hospital, LMU Munich, in Germany, and colleagues performed machine learning-based differentiation of benign and premalignant polyps detected with CT colonography in a sample from an asymptomatic colorectal cancer screening population at average risk. Colorectal polyps were manually segmented on CT colonographic images and were classified as benign or premalignant. After applying 22 image filters, quantitative image features characterizing shape, gray-level histogram statistics, and image texture were extracted from segmentations, resulting in 1,906 feature-filter combinations.

The random forest model was fitted using a training set with 107 colorectal polyps in 63 patients comprising 169 segmentations on CT colonographic images, while the external test set included 77 polyps from 59 patients comprising 118 segmentations. The researchers found that the area under the receiver operating characteristic curve was 0.91 in a random forest analysis, with sensitivity and specificity of 82 and 85 percent, respectively, in the external test set. In subgroup analyses of two size categories, the area under the receiver operating characteristic curve was 0.87 and 0.90 for size categories 6 to 9 mm and 10 mm or larger, respectively. Quantifying first-order gray-level histogram statistics was the most important image feature for decision-making.

“Adding machine learning-assisted image analysis to conventional radiologic image reading could further improve the clinical importance of CT colonography-based colorectal cancer screening by allowing for a more precise selection of patients eligible for subsequent optical colonoscopy-guided polypectomy,” the authors write.

Abstract/Full Text

Copyright © 2020 HealthDay. All rights reserved.