Diagnosis of Nonperitonealized Colorectal Cancer with Computerized Tomography Image Features under Deep Learning

Abstract

This study aimed to explore the value of abdominal computerized tomography (CT) three-dimensional reconstruction using the dense residual single-axis super-resolution algorithm in the diagnosis of nonperitonealized colorectal cancer (CC). 103 patients with nonperitonealized CC (the lesion was located in the ascending colon or descending colon) were taken as the research subjects. The imagological tumor (T) staging, the extramural depth (EMD) of the cancer tissues, and the extramural vascular invasion (EMVI) grading were analyzed. A dense residual single-axis super-resolution network model was also constructed for enhancing CT images. It was found that the CT images processed using the algorithm were clear, and the peak signal-to-noise ratio (PSNR) and structural similarity (SSIM) were 33.828 dB and 0.856, respectively. In the imagological T staging of CC patients, there were 17 cases in the T3 stage and 68 cases in the T4 stage. With the EMD increasing, the preoperative carcinoembryonic antigen (CEA) highly increased, and the difference was statistically significant (P < 0.05). The postoperative hospital stays of patients were also different with different grades of EMVI. The hospital stay of grade 1 patients (19.45 days) was much longer than that of grade 2 patients (13.19 days), grade 3 patients (15.36 days), and grade 4 patients (14.36 days); the differences were of statistical significance (P < 0.05). It was suggested that CT images under the deep learning algorithm had a high clinical value in the evaluation of T staging, EMD, and EMVI for the diagnosis of CC.