Predicting the pathological subdiagnosis of benign prostatic hyperplasia with MRI radiomics: A noninvasive approach

Abstract To predict the pathological subdiagnosis of benign prostatic hyperplasia (BPH‐PS) with the overarching aim of mitigating finasteride overtreatment, we conducted a study integrating clinical, pathological, and radiomic data from BPH patients at our center. The study comprised a retrospective analysis of data and a prospective cohort of BPH patients undergoing pharmacological treatment for six months. Sixty patients (mean age: 68.5 ± 7.1 years) with 360 histopathological whole‐slide images (WSIs) were included in the retrospective cohort. Three gland recognition models were developed and trained to identify glandular structures in WSIs, facilitating the calculation of gland ratios. The optimal model, based on Artificial Neural Network—Multilayer Perceptron (ANN‐MLP), achieved an intraclass correlation coefficient (ICC) of .948 ( p < .001) compared to manual annotation. Radiomic features and habitat analysis were extracted from apparent diffusion coefficient (ADC). Patients were categorized into two groups, BPH with predominantly glandular hyperplasia (BPH‐G) and BPH with predominantly stromal hyperplasia (BPH‐S). An optimal ADC cut‐off value of 1.110×10 −3 mm 2 /s was determined to calculate the Habitat Score. The Habitat Score demonstrated superior predictive performance for BPH‐PS, with an area under the curve (AUC) value of .902. A total of 161 patients (mean age: 64.3 ± 7.0 years) were enrolled in a prospective cohort study. By comparing the changes in prostate volume under MRI before and after medication, it was found that the volume reduction rate of BPH‐G was significantly higher than that of BPH‐S ( p < .001). In conclusion, the ADC value was associated with the gland ratio and could be utilized to predict BPH‐PS, which suggested that the therapeutic efficacy of finasteride could be anticipated by diagnosing BPH‐PS.