Quantification of intratumoral heterogeneity using habitat-based MRI radiomics for predicting high-Gleason scores and castration-resistant PCa: retrospective study

Prostate cancer (PCa) with high-Gleason Scores (GS) tends to have aggressive clinicopathological characteristics and a poor prognosis. we aimed to develop and validate an MRI-based habitat imaging (HI) model for the preoperative prediction of high-GS and castration-resistant prostate cancer (CRPC).

We collected T2-weighted imaging (T2WI), diffusion-weighted imaging (DWI), and apparent diffusion coefficient (ADC) images from 264 prostate cancer patients.Task 1 involved distinguishing between high and low GS PCa, followed by Task 2 which aimed to predict the development of CRPC in high-GS PCa cases. Task 1 extracted whole-tumor radiomic features and tumor microenvironment heterogeneity-based radiomic features from MRI images of 264 patients to construct a radiomic signature and an intratumoral heterogeneity (ITH) signature. Multivariate logistic regression analysis was employed to identify significant independent clinical predictive variables, which were then integrated with the radiomic and ITH signatures to develop a combined model. In Task 2, whole-tumor and tumor microenvironment heterogeneity-based radiomic features were extracted from MRI images of 142 high-GS patients to build an intratumoral heterogeneity signature. The discriminatory performance of these features was evaluated through receiver operating characteristic (ROC) curve analysis, while subsequent decision curve analysis (DCA) was conducted to assess the clinical utility value of the radiomics features.

In Task 1, the cohort of 264 patients was randomly split into a training set (n=184) and a validation set (n=80), corresponding to a ratio of 7:3. The ITH model demonstrated superior performance over the Radiomics model in both the training and test sets, with AUC values of 0.892 and 0.826, respectively, for predicting high-GS PCa. The combined model achieved even better performance, yielding AUCs of 0.900 and 0.832 in the training and test sets. In Task 2, among the 169 enrolled cases with GS ≥ 4+3, 142 were successfully followed up, of which 25 developed CRPC within one year after ADT. The ITH model constructed for this task achieved optimal AUC values of 0.802 and 0.840 in the training and test sets, respectively.

Our study developed ITH signatures that quantified ITH using habitat-based MRI radiomics, which demonstrated outstanding performance in predicting both high GS and CRPC.