Texture analysis (TA) of computed tomography (CT) scan images may provide useful quantitative information that can aid in differentiating between colorectal cancer (CRC), colonic lesions from inflammatory bowel disease (IBD), and normal thickened colon wall (NTC) in patients undergoing CT scans, according to results from a retrospective study published in Scientific Reports.
Researchers analyzed data from multiphase CT scans (pre-contrast, arterial, and portal venous phases) performed in 107 patients with CRC, 113 with colonic lesions caused by IBD, and 96 with NTC in Shanghai, China between 2014 and 2018. All disease states were confirmed histologically. Researchers compared classifications by TA with visual classifications made by radiologists (4 gastroenterological radiologists with 3, 4, 10, and 12 years of experience each) using a nonlinear discriminant analysis.
The misclassification rate for TA was low, ranging from 1.66% to 31.84%. Neither CT phase nor feature selection were significantly different. TA better distinguished between 2 tissue types than distinguishing 3 tissues at once, with misclassification rates ranging from 1.66% to 14.04% and 12.61% to 31.84% for each mode of comparison, respectively.
The highest reported accuracies of TA for all analysis combinations were 94.3% when distinguishing between CRC and IBD, 98.0% when distinguishing between CRC and NTC, 93.1% when distinguishing between IBD and NTC, and 81.4% for the 3-way comparison. The researchers’ analysis was able to distinguish tumors from non-tumors in 81.4% of images.
One limitation of this study is that the researchers chose to use 2-dimentional, as opposed to 3-dimentional, TA. Previous studies have concluded that a 3-dimentional TA is the more accurate of the 2, though it is unclear how these results correlate with or contradict with the findings of the present study.
Based on the observed error rates, the study authors contend that useful quantitative information is available in CT images, and by using TA, differentiation between patients with CRC, lesions from IBD, and NTC was possible.
Wang X, Yuan M, Mi H, et al. The feasibility of differentiating colorectal cancer from normal and inflammatory thickening colon wall using CT texture analysis. Sci Reports. 2020;10:6346. doi:10.1038/s41598-020-62973-1.