Clinical-Radiomics Model Based on T2-Weighted and Diffusion-Weighted Magnetic Resonance Images on Predicting "Double Gray Zone" Prostate Cancer of PSA and PI-RADS.
Authors
Affiliations (6)
Affiliations (6)
- First Clinical Medical College, Shanxi Medical University, Taiyuan, China (W.Z., J.L., K.S., Z.G.); Department of Urology, The First Hospital of Shanxi Medical University, Taiyuan, China (W.Z., J.L., X.C., W.S.).
- Department of Radiology, The First Hospital of Shanxi Medical University, Taiyuan, Shanxi, China (X.Y., H.Z.).
- First Clinical Medical College, Shanxi Medical University, Taiyuan, China (W.Z., J.L., K.S., Z.G.); Department of Pathology, The First Hospital of Shanxi Medical University, Taiyuan, China (K.S.).
- Department of Urology, The First Hospital of Shanxi Medical University, Taiyuan, China (W.Z., J.L., X.C., W.S.).
- Department of Radiology, The First Hospital of Shanxi Medical University, Taiyuan, Shanxi, China (X.Y., H.Z.); Intelligent Imaging Big Data and Functional Nano-Imaging Engineering Research Center of Shanxi Province, First Hospital of Shanxi Medical University, Taiyuan, China (H.Z.).
- Department of Urology, The First Hospital of Shanxi Medical University, Taiyuan, China (W.Z., J.L., X.C., W.S.). Electronic address: [email protected].
Abstract
This study aimed to develop a clinical-radiomics model based on T2-weighted imaging (T2WI) and diffusion-weighted imaging (DWI) to improve the diagnostic performance in the "double gray zone" population characterized by prostate-specific antigen (PSA) levels of 4-10 ng/mL and a Prostate Imaging Reporting and Data System (PI-RADS) score of 3. A total of 93 eligible patients (PSA 4-10 ng/mL and PI-RADS score 3) were retrospectively included. Radiomics features were extracted from T2WI and DWI sequences. Feature selection was performed using Maximum Relevance Minimum Redundancy (mRMR) and Least Absolute Shrinkage and Selection Operator (LASSO) regression. Six machine learning algorithms were used to construct the T2WI, DWI, DWI_T2, and clinical models and the logistic regression (LR) algorithm was used to construct a combined model. 10 T2WI and 13 DWI features were selected to build radiomics models. Among the radiomics models, the DWI_T2 fusion model demonstrated improved performance compared to the other two models, and the ExtraTrees algorithm performed best, with the area under the curve (AUC) value of 0.833 (95% confidence interval (CI): 0.662-1.000) and an F1 score of 0.778 in the validation set. Key clinical variables-age, Upper_and_lower_diameter, and prostate-specific antigen density (PSAD)-were identified via LASSO. Meanwhile, the clinical model built using RandomForest achieved a validation set AUC of 0.767 (95% CI: 0.561-0.973). A combined clinical-radiomics model was further constructed, achieving a validation set AUC of 0.867, outperforming both the standalone clinical model and the standalone radiomics model. Calibration curves, the Hosmer-Lemeshow (HL) test, and decision curve analysis (DCA) confirmed that the combined model demonstrated good fit and clinical net benefit. The combined clinical-radiomic model we constructed demonstrate good diagnostic performance in the "double gray zone" population, holding potential as an effective non-invasive tool to aid clinical decision-making.