Biparametric magnetic resonance imaging-based radiomics model can improve the detection of dense and sparse prostate cancers.

Background: Prostate cancer (PCa) tissues are heterogeneous. The tumor lesion could be either the dense focus consisting of high proportion of malignant glands or the sparse focus consisting of a mixture of scattered tumor glands and normal tissue. The different growth patterns will affect the detection of PCa. The detection rate of the sparse PCa is far lower than that of the dense PCa. This study aimed to explore the value of radiomics based on biparametric magnetic resonance imaging (bpMRI) in the detection of dense and sparse PCa lesions. Methods: A total of 372 PCa lesions of 156 patients from two centers were defined as "sparse" or "dense" according to whole-mount sections and then delineated on bpMR images. For each lesion, 2,553 radiomics features were extracted from images. The optimal radiomics features were selected by one-way analysis of variance (ANOVA) and least absolute shrinkage and selection operator (LASSO) regression. The radiomics models constructed by the random forest classifier and the average apparent diffusion coefficient (ADC) value model were established for the peripheral zone (PZ) and the transitional zone (TZ) to detect dense lesions, sparse lesions, and noncancerous tissues. The areas under the curve (AUCs) and DeLong tests were used to analyze the performance of the models. Results: In the PZ, the AUCs of the radiomics model (external validation set) for noncancerous tissue, dense lesion, and sparse lesion detection were 0.91, 0.98, and 0.88, respectively, and those of the average ADC value model were 0.85, 0.73, and 0.62, respectively. In the TZ, those of the radiomics model were 0.92, 0.93, and 0.88, respectively, and those of the average ADC value model were 0.81, 0.83, and 0.52, respectively. Compared with that of the average ADC value model, the AUC of the radiomics model in the diagnosis of sparse lesions significantly differed (P<0.05). In the detection of dense lesions, there were significant differences in the training set (P<0.05). Conclusions: The radiomics model based on bpMRI can effectively improve the detection of PCa lesions, especially sparse lesions, and significantly reduce missed diagnoses.