Texture analysis of CT colonography to develop a novel imaging biomarker for the management of colorectal cancer

Abstract

AbstractBackgroundRecent studies have focused on evaluating the biomarker value of textural features in radiological images. Our study investigated whether or not a texture analysis of computed tomographic colonography (CTC) images could be a novel biomarker for colorectal cancer (CRC).MethodsThis retrospective study investigated 263 patients with CRC who underwent contrast‐enhanced CTC (CE‐CTC) before curative surgery between January 2014 and December 2017. Multiple texture analyses (fractal, histogram, and gray‐level co‐occurrence matrix [GLCM] texture analyses) were applied to CE‐CTC (portal‐venous phase), and fractal dimension (FD), skewness, kurtosis, entropy, and GLCM texture parameters, including GLCM‐correlation, GLCM‐autocorrelation, GLCM‐entropy, and GLCM‐homogeneity, of the tumor were calculated. These texture parameters were compared with pathological factors (tumor depth, lymph node metastasis, vascular invasion, and lymphatic invasion) and overall survival (OS).ResultsTumor depth was significantly associated with FD, kurtosis, entropy, GLCM‐correlation, GLCM‐autocorrelation, GLCM‐entropy, and GLCM‐homogeneity (p = 0.001, 0.001, 0.001, 0.001, 0.018, 0.008, and 0.001, respectively); lymph node metastasis was associated with GLCM‐homogeneity (p = 0.004); lymphatic invasion was associated with GLCM‐correlation and GLCM‐homogeneity (p = 0.001 and 0.012, respectively); and venous invasion was associated with FD, entropy, GLCM‐correlation, GLCM‐autocorrelation, and GLCM‐entropy of the tumor (p = 0.001, 0.033, 0.021, 0.046, respectively). In the Kaplan–Meier analysis, patients with high GLCM‐correlation tumors or high GLCM‐homogeneity tumors showed a significantly worse OS than others (p = 0.001 and 0.04, respectively). Multivariate analyses showed that the GLCM correlation was an independent prognostic factor for the OS (p = 0.021).ConclusionCE‐CTC‐derived texture parameters may be clinically useful biomarkers for managing CRC patients.