Magnetic resonance imaging radiomics-based prediction of clinically significant prostate cancer in equivocal PI-RADS 3 lesions in the transitional zone

• DOI: 10.3389/fonc.2023.1247682
• Publication Date: 2023-11-23T11:46:04Z
• Authors: Ying-Ying Zhao; Mei-Lian Xiong; Yue-Feng Liu; Li-Juan Duan; Jia-Li Chen; Zhen Xing; Yan-Shun Lin; Tan-Hui Chen
• Keywords: Univariate; Univariate analysis; Nomogram

Purpose This bi-institutional study aimed to establish a robust model for predicting clinically significant prostate cancer (csPCa) (pathological grade group ≥ 2) in PI-RADS 3 lesions the transition zone by comparing performance of combination models. Materials and methods included 243 consecutive men who underwent 3-Tesla magnetic resonance imaging (MRI) ultrasound-guided transrectal biopsy from January 2020 April 2022 which is divided into training cohort 170 patients separate testing 73 patients. T2WI DWI images were manually segmented mean ADC radiomic analysis. Predictive clinical factors identified using both univariate multivariate logistic The least absolute shrinkage selection operator (LASSO) regression models deployed feature constructing signatures. We developed nine utilizing factors, radiological features, radiomics, leveraging XGboost methods. performances these was subsequently compared Receiver Operating Characteristic (ROC) analysis Delong test. Results Out participants with median age 70 years, 30 diagnosed csPCa, leaving 213 without csPCa diagnosis. Prostate-specific antigen density (PSAD) stood out as only factor (odds ratio [OR], 1.068; 95% confidence interval [CI], 1.029–1.115), discovered through Seven features correlated prediction. Notably, outperformed eight other (AUC cohort: 0.949, validation 0.913). However, it did not surpass PSAD+MADC (P > 0.05) cohorts (AUC, 0.949 vs. 0.888 0.913 0.854, respectively). Conclusion machine learning presented best within transitional zone. addition classifiers display any enhancement over compound findings. most exemplary generalized option quantitative evaluation Mean ADC+PSAD.