Multimodal Nomogram Combining Multiparametric MRI, Functional Subsets of Peripheral Lymphocytes and PI-RADS Can...

Risk stratification is a crucial aspect for clinical decision-making and prognostic prediction for patients with prostate cancer (PCa). Noninvasive tripartite risk stratification, which classifies PCa patients into low, intermediate, and high-risk groups, is highly informative for guiding treatment selection, but has rarely been explored. Classic PSA-based risk stratification of PCa has raised serious doubts about its accuracy. We aimed to investigate alternatives to PSA for performing automated and reliable risk stratification.

This retrospective study enrolled 110 patients who underwent MRI scans and peripheral lymphocyte examinations at two campuses of Wuhan Tongji Hospital between August 1st, 2020, and October 20th, 2022. Two experienced radiologists reviewed the multiparametric MRI (mpMPI) images in accordance with the Prostate Imaging and Reporting and Data System (PI-RADS) v2.1 guidelines. A multimodal nomogram was developed to predict tripartite risk stratification in PCa patients by integrating mpMRI, functional subsets of peripheral lymphocytes, and clinical variables such as PI-RADS scoring, prostate volume, and age. Both radiomics and deep learning classification models were employed in constructing the multimodal nomogram. Its performance was compared against that of a traditional clinical nomogram and independent predictors.

The multimodal nomogram demonstrated superior performance, achieving an AUC value of 0.9609 (CI 0.8798-1.0000), an F1 score of 0.8671 (CI 0.6983-1.0000), a sensitivity of 0.9333(CI 0.8667-1.0000), and a specificity of 0.8667 (CI 0.7000-1.0000) in the test set. The predictive scores derived by the nomogram displayed strong correlations with PCa risk stratifications. The multimodal nomogram significantly outperformed the clinical nomogram and various predictors including PSA.

The proposed nomogram indicates its potential as an alternative diagnostic tool for reliably categorizing PCa patients into low-, intermediate-, and high-risk groups. Such an automated tool could help reduce unnecessary biopsies and identify candidate patients who may be more suitable for active surveillance or radical prostatectomy.