Primary Exploration Of Biparametric-MRI 3D Texture Analysis For Detecting And Evaluating High-grade Prostate Cancer

Objectives: To investigate the performance of three-dimensional texture analysis of Biparametric MR imaging in the diagnosis and assessment of high-grade prostate cancer.Methods: A retrospective research of 116 patients with clinically suspected prostate cancer undergoing multiple-parameter magnetic resonance and histopathological biopsies was retrospectively studied.Magnetic resonance images were scored by two radiologists using second version of the Prostate Imaging Reporting Data System with double-bling method,and calculated the texture analysis parameters divided by biparametric MRI.The PI-RADS v2 score and texture analysis parameters were evaluated between different observers' Uniformity.A logistic regression model of different regions of the prostate based on texture analysis parameters was established.Use ROC curve analysis to compare models and other parameters(kurtosis,skewness,energy,entropy,correlation and inertia)based on texture analysis parameters.The correlation between the high-grade prostate cancer Gleason score and each parameter was assessed.Results: The PI-RADS v2 score had moderate to good reliability(All lesions: intraclass correlation coefficients in transition zone and peripheral zone were 0.699-0.915 and 0.813-0.894,respectively;High-grade cancer: intraclass correlation coefficients in transition zone and peripheral zone were 0.836-0.960 and 0.862-0.928,respectively),and the texture analysis parameters had good to excellent reliability(All lesions: intraclass correlation coefficients in transition zone and peripheral zone were 0.747 and 0.728,respectively;High-grade cancer: intraclass correlation coefficients in transition zone and peripheral zone were 0.709 and 0.963,respectively).Multivariate logistic regression analysis showed that the skewness,energy and inertia parameters of high-grade cancer in transition zone were independent predictor.And in peripheral zone,the skewness,correlation and inertia parameters were independent predictors.Compared with PI-RADS v2 and other texture parameters(kurtosis,skewness,energy,entropy,correlation and inertia),the diagnostic performance based on the texture analysis parameter model was significantly improved(The area under the curve of transition zone and peripheral zone were 0.990 and 0.963,respectively).For transition zone and peripheral zone,entropy showed moderate correlation with GS of HGPCa(r=0.441,P=0.031;r=0.355,P=0.046,respectively).While,inertia had negative correlations with GS(r=-0.448,P=0.010)in peripheral zone.Conclusions: This clinical Study has shown that texture analysis-based parametric models established by biparametric MR imaging can be used to identify high-grade prostate cancer.Meanwhile,specific parameters(entropy and inertia)extracted by texture analysis may provide trusted tools for tumor aggressiveness assessment.