Non-invasive MRI-based assessment of reactive stromal grade in prostate cancer using diffusion kurtosis imaging and stretched-exponential model.

Abstract

Objectives: Reactive stroma plays a pivotal role in the genesis, progression, and metastasis of prostate cancer (PCa). Higher reactive stromal grade (RSG) generally portends a poorer prognosis. The aim of the study is non-invasively evaluate RSG by preoperative mono-exponential model, stretch-exponent model (SEM) and diffusion kurtosis imaging (DKI), and isolate the independent predictor of high RSG (> 50% reactive stroma) in parameters of mono-exponential model, SEM and DKI.

Methods: Totally, 54 low RSG (≤ 50% reactive stroma) patients and 26 high RSG patients were prospectively enrolled in the study. Apparent diffusion coefficient (ADC), mean kurtosis (MK), mean diffusivity (MD), distributed diffusion coefficient (DDC), and heterogeneity index (α) values of all lesions were measured on GE Workstation 4.6. Spearman's rank correlation analysis was used to analysis the correlation between RSG and parameters of SEM and DKI. Receiver-operating characteristic (ROC) curve analysis was performed to evaluate the diagnostic performance of those parameters in differentiating low RSG and high RSG. DeLong's test was used to assess whether the differences of AUC for each parameter were statistically significant. Binary logistic regression analysis was performed to identify independent predictors of high RSG.

Results: ADC (r = - 0.352, p = 0.001), DDC (r = - 0.579, p < 0.001) and MD (r = - 0.597, p < 0.001) values showed significant negative correlations with RSG, while MK value (r = 0.658, p < 0.001) demonstrated a significant positive correlation. MK (AUC = 0.816, p < 0.001) was superior to ADC (AUC = 0.717, p < 0.001), DDC (AUC = 0.781, p < 0.001) and MD (AUC = 0.774, p < 0.001) in differentiating low and high RSG, but the differences between these AUCs were not statistically significant (all p > 0.05). Binary logistic regression analysis demonstrated a statistically significant model (χ² =43.222, p < 0.001), and showed that MK (odds ratio = 10.185; 95% CI: 2.467 ~ 21.694; p < 0.001) and MD (odds ratio = 0.014; 95% CI: 0.003 ~ 0.367; p < 0.001) were the independent predictors of high RSG.

Conclusion: Although ADC, DDC, and MD values were significantly negatively correlated with RSG, and MK was significantly positively correlated, and all three models-mono-exponential model, SEM, and DKI-demonstrated good performance in differentiating between low and high RSG, only parameters MD and MK values of DKI were identified as independent predictors of high RSG.