Towards the Definition of Radiomic Features and Clinical Indices to Enhance the Diagnosis of Clinically Significant Cancers in PI-RADS 4 and 5 Lesions

Abstract: Objectives: Lesions classified as PI-RADS 4/5, according to the Prostate Imaging–Reporting and Data System (PI-RADS) guidelines, may include false positives. This study aims to identify promising radiomic features that may support the detection of clinically significant tumours among PI-RADS 4/5 lesions on MRI. Methods: Patients undergoing a 3T magnet multiparametric MRI (mpMRI) for clinical suspicion of prostate cancer (PC) or active surveillance were retrospectively enrolled. Pathological results utilizing MRI-targeted biopsy specimens were considered the ground truth. Clinical (age, PSA, PSA density) and MRI parameters (prostate volume, mean apparent diffusion coefficient/ADC) were collected. Lesions were manually contoured on axial T2-weighted images and ADC maps. Radiomic features were extracted with Pyradiomics. Clinical and radiomic features best correlating with histopathological results were selected. Diagnostic values were assessed on validation samples. Results: The final cohort included 99 patients (mean age, 69.2 ± 6.8 years) and 111 PI-RADS 4/5 lesions. At pathology, 79 lesions (71%) were identified as clinically significant cancers (Gleason score ≥ 7). Radiomic, clinical, and MRI features best correlating with histopathology were selected. The best predictive clinical and radiomic multivariate model showed the following diagnostic values: sensitivity, 79%; specificity, 80%; positive predictive value (PPV), 91%; negative predictive value (NPV), 63%; accuracy, 79%. A radiomic multivariate model based exclusively on peripheral zone lesions showed more promising values: sensitivity, 86%; specificity, 80%; PPV, 93%; NPV, 70%; accuracy, 84%. Conclusions: Radiomic MRI feature analysis can potentially improve the accuracy of mpMRI in discriminating between clinically significant cancers in PI-RADS 4 and 5 lesions.