Predicting tumor deposits in patients with rectal cancer: Using the models of multiple mathematical parameters derived from diffusion-weighted imaging

Using mono-exponential, bi-exponential, and stretched-exponential models of multi-b-value diffusion-weighted imaging (DWI) to predict tumor depositions (TDs) in patients with rectal cancer (RC).This retrospective study, between January 2018 and November 2021, enrolled 30 TDs-positive and 38 TDs-negative of patients with rectal cancer. The mathematical parameters including ADC from mono-exponential model, D, D* and f from bi-exponential model, and DDC and α from stretched-exponential model, clinical factors (such as age, gender, pathological stage, etc.) and image features (such as length, thickness, location, etc.) from tumor characteristics were obtained to identify the two groups of patients. The results were evaluated by the receiver operating characteristic curve (ROC) analysis and area under the ROC curve (AUC). Multivariate binary logistic regression analysis was conducted to determine the independent risk factors.The D* and α values, pt. stage, tumor location, mesorecta fascia (MRF) / peritoneum status and percentage of rectal wall circumference invaded (PCI) were significantly different between the TDs-positive and TDs-negative groups (P < 0.001, P < 0.001, P = 0.029, P = 0.008, P < 0.001 and P = 0.002, respectively), with the AUC were 0.838, 0.901, 0.618, 0.698 0.694 and 0.758, respectively. The D* and α values were proved to be independent risk factors after multivariate binary logistic regression analysis (p = 0.022 and 0.004, respectively). The AUC of the model consisting of the D* and α values was 0.913 (95 % CI 0.820 ∼ 0.968).The combined model constructed by D*, α and tumor location demonstrated superior diagnostic performance, with the AUC, sensitivity, specificity, and accuracy of 0.947 (95 % confidence interval, CI, 0.865-0.987), 0.900, 0.868 and 0.853, respectively.Multiple mathematical parameters can be used as preoperative auxiliary diagnostic tools to predict TDs of RC. The combined model constructed by D*, α and tumor location show better diagnostic performance for TDs.