Assessment of invasion of prostate cancer with multiparametric MRI-based radiomics

Abstract

Objective To investigate the value of a radiomic model of multiparametric MRI different regions of interest (ROI) in assessment of the invasion of prostate cancer (PCa), and explore the evaluation value of an integrated model based on above radiomic model combined with radiomics, PI-RADS 2.1 score, and clinical variables. Methods A total of 245 patients with pathologically-confirmed PCa admitted to 2 medical centers in our hospital from May 2018 to September 2022 were collected in this retrospective study. Among them, 176 cases were collected in Yuzhong Medical Center, including 77 cases in the low invasive group [Gleason score ≤7 (3+4)] and 99 cases in the high invasive group [Gleason score ≥7 (4+3)]; 69 cases were collected in Jiangnan Medical Center, including 33 cases in the low invasive group and 36 cases in the high invasive group. All patients underwent multiparametric MRI, and then 2 types of ROI, the tumor region (TR) and the prostate gland (PG), were segmented on the multiparametric MRI images. Clinical variables related with PCa invasion were assessed, and PI-RADS 2.1 score was recorded for each patient. Logistic regression algorithm was employed as a machine learning method to develop following invasive stratification models for PCa: radiomics models (ModelTR, ModelPG and ModelPG+TR), Radiomics-Clinical model, Radiomics-PIRADS model, PIRADS-Clinical model and Radiomics-PIRADS-Clinical combined model. Receiver operating characteristic (ROC) curve, area under ROC curve (AUC) and decision curve analysis (DCA) were applied to compare the diagnostic efficacy and clinical benefit of each model. A nomogram was constructed by combining Radiomics score (Radscore), PI-RADS 2.1 score and independent clinical variables, and its performance was evaluated by calibration, differentiation and clinical application. Results In the above 3 radiomics models, the AUC value was 0.919 for ModelPG+TR, which was higher than that of ModelTR (0.874) and of ModelPG (0.887). And the AUC value of the Radiomics-PIRADS-Clinical combined model was 0.954, superior to that of the radiomics model (0.919), Radiomics-PIRADS model (0.921), Radiomics-Clinical model (0.919), and PIRADS-Clinical model (0.769) respectively, The nomogram model showed good performance (AUC=0.919) and calibration efficacy in risk stratification. DSA revealed that both the ModelPG+TR model and Radiomics-PIRADS-Clinical combined models achieved a higher net clinical benefit. Conclusion Our radiomic model of combining features of the prostate and tumor regions can more accurately assess the invasion of PCa, and our integrated model of combining radiomics, PI-RADS 2.1 score and clinical variables can further improve the assessment.

Keywords

• radiomics
• multiparametric magnetic resonance imaging
• prostate cancer
• invasion
• prostate volume
• tumor volume