A two-stage model for precise identification and Gleason grading of clinically significant prostate cancer: a hybrid approach

Abstract

Introduction: Accurate identification and grading of clinically significant prostate cancer (csPCa, Gleason Score ≥ 7) without invasive procedures remains a significant clinical challenge. This study aims to develop and evaluate a two-stage model designed for precise Gleason grading. The model initially uses radiomics-based multiparametric MRI to identify csPCa and then refines the Gleason grading by integrating clinical indicators and radiomics features.

Methods: We retrospectively analysed 399 patients with PI-RADS ≥ 3 lesions, categorising them into non-significant prostate cancer (nsPCa, 263 cases) and csPCa (136 cases, subdivided by GGs). Regions of interest (ROIs) for the prostate and lesions were manually delineated on T2-weighted and apparent diffusion coefficient (ADC) images, followed by the extraction of radiomics features. A two-stage model was developed: the first stage identifies csPCa using radiomics-based MRI, and the second integrates clinical indicators for Gleason grading. Model efficacy was evaluated by sensitivity, specificity, accuracy and area under the curve (AUC), with external validation on 100 patients.

Results: The first-stage model demonstrated excellent diagnostic accuracy for csPCa, achieving AUCs of 0.989, 0.982 and 0.976 in the training, testing and external validation cohorts, respectively. The second-stage model exhibited commendable Gleason grading capabilities, with AUCs of 0.82, 0.844 and 0.83 across the same cohorts. Decision curve analysis supported the clinical applicability of both models.

Conclusions: This study validated the potential of T2W and ADC image radiomics features as biomarkers in distinguishing csPCa. Combining these features with clinical indicators for csPCa Gleason grading provides superior predictive performance and significant clinical benefit.

Keywords: Clinically significant prostate cancer (csPCa); Gleason grading; multiparametric MRI (mpMRI); prostate cancer (PCa); radiomics.

Figure 1

Flow diagram of patient selection for the study. Institution 1: Xiaogan Hospital Affiliated to Wuhan University of Science and Technology. Institution 2: Tongji Hospital, Tongji Medical College, Huazhong University of Science and Technology.

Figure 2

The Pipeline of radiomics analysis for Gleason Grading in Prostate Cancer Patients. (A) Image acquisition and segmentation. (B) Feature extraction. (C) Feature selection. (D) Feature fusion. (E) Model construction.

Figure 3

The ROC curve, calibration curve and decision curve of the multimodal feature fusion radiomics models for the classification of csPCa and nsPCa. Panels (A, D, G) represent the training set; panels (B, E, H) represent the testing set; and panels (C, F, I) represent the external validation set.

Figure 4

The ROC curve, calibration curve and decision curve of the multimodal feature fusion radiomics models (T2W_ADC + clinical) for the classification of Gleason grade (2, 3/4, 5). Panels (A, D, G) represent the training set; panels (B, E, H) represent the testing set; and panels (C, F, I) represent the external validation set.