Predictive modelling for prostate cancer aggressiveness using non-invasive MRI techniques

Magnetic Resonance Imaging (MRI) plays a crucial role in the diagnosis of prostate cancer (Pca). This study aimed to improve the diagnostic accuracy of MRI in distinguishing between prostate tumours of Grade Group (GG)2 versus GGs3-5 and GG2 versus GG3 only, using predictive models.

Double Inversion Recovery MRI (DIR-MRI) and Multiparametric MRI (mpMRI) scans from 53 patients (mean age: 67 years) acquired between January 2015 and January 2017 were retrospectively analysed. The suspected PCa lesions identified on MRI were correlated with biopsy targets and GGs. Lesion-to-normal ratios (LNRs) of potential PCa lesions were calculated using the Siemens Healthineers Syngo.via Picture Archiving and Communication System (PACS) by drawing Regions of Interest (ROIs) around the lesions and corresponding normal tissue to measure their respective signal intensities. Prediction models were developed using the R statistical package CARRoT, integrating MRI-derived variables and baseline patient characteristics to reliably classify PCa GGs.

The developed predictive models achieved high diagnostic performance, with Area Under the Receiver Operating Characteristic Curve (AUROC) of 0.86 and 0.91 upon 1000 cross-validations, respectively.

We present explainable and rigorously cross-validated models that differentiate less aggressive from more aggressive PCa based on T2 LNR and the tumuor short axis measured on axial T2-weighted MRI (Dimension B). In contrast to existing models, which often lack validation (internal or external) or rely on non-explainable Artificial Intelligence techniques, our models offer greater clinical applicability.

These models provide a robust, explainable tool for clinicians to accurately distinguish between less and more aggressive PCa, utilizing T2 LNR and axial T2 tumuor dimensions. By addressing limitations in existing predictive models, they offer potential for improved clinical decision-making.