Salivary gland carcinoma: Prediction of cancer death risk based on apparent diffusion coefficient histogram profiles.

We evaluated apparent diffusion coefficient (ADC) histogram parameters for predicting the outcomes of patients with salivary gland carcinoma. Diffusion-weighted MR imaging was performed in 20 patients with salivary gland carcinoma, and ADCs were determined using b-values of 500 and 1000 s/mm2. ADC histogram parameters (mean, median, percentage tumor area with distinctive ADC values [pADC], skewness, and kurtosis) were analyzed. The patients were followed for 5-136 months after primary surgery. The ADC histogram parameters and T (pT), N(pN), and M categories of the primary tumors were assessed for the prognostic importance using Cox proportional hazards models, logistic regression analysis, and receiver operating characteristic (ROC) analysis. Cohen's d was determined for evaluating the importance of differences in the parameters between two patient groups with different outcomes. Six patients died of cancer (DOC) within 3 years after the primary surgery. Cox proportional hazards models indicated that ADC mean (95% CI = 0.494-0.977, p = 0.034), ADC median (95% CI = 0.511-0.997, p = 0.048), pADC with extremely low (<0.6 mm2/s) ADC (95% CI = 1.013-1.082, p = 0.007), kurtosis (95% CI = 1.166-7.420, p = 0.023), and pN classification (95% CI = 1.196-4.836, p = 0.012) were important factors of cancer death risk. ROC analyses indicated that the pADC <0.6 ×10(-3) mm2/s was the best prognostic predictor (p <0.001; AUC = 0.929) among the ADC and TNM classification parameters that were significant in a univariate logistic regression analysis. Cohen's d values between the DOC and survived patients for the ADC mean, ADC median, pADC with extremely low ADC, and kurtosis were 1.06, 1.04, 2.12, and 1.13, respectively. These results suggest that ADC histogram analysis may be helpful for predicting the outcomes of patients with salivary gland carcinoma.