Statistics

Statistical analysis was performed using Statistical Package for the Social Sciences (SPSS, version 20.0, IBM, Inc., Chicago, IL) and MedCalc Statistical Software (version 18.2, Ostend, Belgium).

Interobserver agreement on each IVIM-MRI parameter for two independent readers was analysed by estimating the intraclass correlation coefficient (ICC). Based on the method given by Cicchetti [20], the ICC values were interpreted as follows: < 0.40, poor inter-rater agreement; 0.40–0.59, fair; 0.60–0.74, good; and 0.75–1.00, excellent. Bland-Altman plots were also constructed, and the limits of agreement (LoAs) were estimated from the plots.

According to the results of the one-sample Shapiro-Wilk test, continuous data were expressed as the means ± standard deviations (SDs) or the medians with interquartile ranges (IQRs). The independent-samples t test and the Mann-Whitney U test were applied for normally and non-normally distributed data, respectively.

Based on the results, we constructed receiver operation characteristic (ROC) curves for the statistically significant parameters derived from IVIM-MRI. Then, the optimum parameters were selected to establish a multi-parameter combined equation using the bi-logistic regression method. The ROC curve of predictive probability was also generated. The areas under the ROC curves (AUCs), with confidence intervals (CIs), were calculated. An AUC value of < 0.5 indicated no diagnostic performance; 0.5–0.7, poor performance; 0.7–0.9, moderate performance; and > 0.9, excellent performance. The maximum Youden indexes were used to determine the respective cutoff values producing the best diagnostic accuracy.

The chi-square test or Fisher’s exact test was used to compare the clinical and pathological features of rectal tumours between node-negative and node-positive patients. A P value of < 0.05 was considered statistically significant.