Texture Analysis of Multiplanar Reconstruction Images from CT Colonography for Differentiating Colorectal Adenomas from Adenocarcinomas

This study aimed to evaluate whether texture analyses applied to multiplanar reconstruction images from computed tomographic colonography can improve differentiation between colorectal adenomas and adenocarcinomas. This retrospective study included 152 patients who underwent computed tomographic colonography for colorectal lesion evaluation at our institution between April 2021 and September 2024. Computed tomographic colonography was performed with and without a contrast medium. All 242 lesions were imaged and analysed on both plain (PCTC) and contrast-enhanced (CECTC) computed tomographic colonography; no lesion was available on only one modality. The two modalities were therefore derived from the same lesion set, and identical denominators (242 lesions: 152 adenomas and 90 adenocarcinomas) were used for each modality. Intramucosal carcinoma was counted as malignant in accordance with the Japanese classification. Protrusions were identified on virtual endoscopic images, and regions of interest were manually delineated on the maximum area of two-dimensional multiplanar reconstruction images. Texture analyses, including histogram analysis, fractal analysis and the gray-level co-occurrence matrices, were applied. These texture parameters were compared between adenomas andadenocarcinomas. Diagnostic performance was assessed using receiver operating characteristic curve analysis. Interobserver reproducibility was evaluated using intraclass correlation coefficients, with all parameters demonstrating ICC ≥ 0.75. A total of 242 lesions (152 adenomas and 90 adenocarcinomas) were included. All texture parameters showed significant differences in differentiation between adenomas and adenocarcinomas on both non-contrast computed tomographic colonography and contrast-enhanced computed tomographic colonography. On plain computed tomographic colonography, GLCM entropy showed the highest diagnostic performance (area under the curve, 0.833; sensitivity, 0.856; specificity, 0.750). On contrast-enhanced computed tomographic colonography, GLCM entropy demonstrated the highest area under the curve (0.811; sensitivity, 0.689; specificity, 0.862). Texture parameters on multiplanar reconstruction images of computed tomographic colonography can effectively distinguish between adenoma and adenocarcinoma, particularly using gray-level co-occurrence matrix entropy. Integrating textural feature analysis into computed tomographic colonography may improve diagnostic performance.