Diagnosis Of Prostate Cancer Using Multiple B-value Diffusion Weighted Imaging And Radiomics Derived From Multiparametric MRI

Chapter One The diagnostic value of multiparametric magnetic resonance imaging which includes multiple b-value diffusion weighted imaging for prostate cancerObjective:To investigate the diagnostic value of multiparametric magnetic resonance imaging（mp MRI）which includes multiple b-value diffusion weighted imaging（DWI）sequence for prostate cancer（PCa）.Develop diagnostic models regarding transition zone and peripheral zone PCa.And make comparison between the diagnostic power of the developed diagnostic models and Prostate Imaging Reporting and Data System Version 2.1（PI-RADS v2.1）.Materials and Methods:Patients undergoing prostate mp MRI in 6 medical centres were retrospectively enrolled.The b-values used in DWI were 50,400,800,1400s/mm².The lesions included in this study were graded according to PI-RADS v2.1 by 2 radiologists who were blinded to the pathology results.Any discrepancies were solved by discussion.The lesions were randomly allocated to training and validation cohort at an approximate ratio of 2:1.The imaging features derived from mp MRI included lenticular shape,T2WI hypointense,indistinct margin,invasion of anterior fibromuscular stroma（AFMS）,early enhancement,prostatic parenchymal diffuse enhancement,and the type of the signal intensity-time（SIT）curve（persistent,plateau,washout）.The diameter and apparent diffusion coefficient（ADC）of the lesions were measured.ADC_50-400,ADC_400-800,ADC_800-1400,ADC_50-1400,ADC_{50-400-800-1400} and f_h were calculated based on multiple b-value DWI using monoexponential model.Univariate and multivariate analysis were conducted for the imaging features and quantitative parameters calculated from multiple b-value DWI.To avoid multicollinearity,only the ADC value with the best overall diagnostic performance was included in multivariate analysis.Factors proven statistically significant in multivariate analysis were selected to develop diagnostic models regarding transition zone and peripheral zone PCa,respectively.The diagnostic power of the models in training and validation cohort was evaluated by plotting receiver operating characteristic（ROC）curve and calculating area under curve（AUC）.Results:166 patients with a total of 190 lesions were included.97 lesions were located in the transition zone,whereas 93 lesions were in the peripheral zone.Indistinct margin,invasion of AFMS,SIT curve type（washout）,diameter,ADC measured,ADC_50-400,ADC_400-800,ADC_800-1400,ADC_50-1400,ADC_50-400-800-1400 and fh showed statistical significance in the univariate analysis of transition zone lesions.ADC_50-400and invasion of AFMS were proven significant in multivariate analysis.The equation of the diagnostic model of transition zone PCa was:logit（P）=15.213+4.268×invasion of AFMS-16.598×ADC_50-400.When used in the training cohort,the model obtained an AUC of 0.963,which was higher than that of the PI-RADS v2.1score（AUC 0.888）.An AUC of 0.879 was achieved in the validation cohort.T2WI hypointense,early enhancement,SIT curve type（plateau,washout）,ADC measured,ADC_50-400,ADC_400-800,ADC_800-1400,ADC_50-1400,ADC_50-400-800-1400 and fh showed statistical significance in the univariate analysis of peripheral zone lesions.The diagnostic model of peripheral zone PCa consisted of f_h and early enhancement,the equation of the model was:logit（P）=-7.245+1.798×early enhancement+9.833×f_h.When used in the training cohort,the model obtained an AUC of 0.878,which was higher than that of the PI-RADS v2.1 score（AUC 0.844）.An AUC of 0.868 was achieved in the validation cohort.Conclusion:The diagnostic models that consist of imaging features derived from mp MRI and quantitative parameters calculated from multiple b-value DWI outperform PI-RADS v2.1 in the diagnosis of PCa.Chapter Two Diagnosis of prostate cancer using multiparametric magnetic resonance imaging: development and validation of a radiomics-based modelObjective: To investigate the value of radiomics features which derived from mp MRI and clinical data in the diagnosis of PCa.Integrate radiomics signature and clinical data to develop a diagnostic model.Materials and Methods:Patients undergoing prostate mp MRI in 6 medical centres were retrospectively enrolled.The b-values used in DWI were 50,400,800,1400s/mm².The lesions included in this study were graded according to PI-RADS v2.1 by 2 radiologists who were blinded to the pathology results.Any discrepancies were solved by discussion.The patients were randomly allocated to training and validation cohort at an approximate ratio of 7:3.The volume of interest（VOI）on T2 WI,DWI（b=1400s/mm²）,ADC map and DCE-MRI were manually delineated.Radiomics features were extracted after VOI delineation.Feature selection was based on L1 regularisation and intraclass correlation coefficient（ICC）analysis.Selected features were normalised for Rad-score calculation.Serum total prostate specific antigen（t PSA）,prostate volume（PV）and prostate specific antigen density（PSAD）of each patient were documented.Univariate analysis was conducted for Radiomics signature and clinical data.Factors showed statistical significance were included in multivariate analysis.And factors proven statistically significant in multivariate analysis were selected to develop the diagnostic model for PCa.A nomogram was plotted to visualise the diagnostic model.And the diagnostic model was calibrated using the bootstrapping method.The diagnostic power of the model in training and validation cohort was evaluated by plotting ROC curve and calculating AUC.Results: A total of 150 patients were included in the study.20 most discriminative radiomics features were selected to calculate the Rad-score.The number of the features extracted from T2 WI,high b-value DWI,ADC map and DCE-MRI were 2,9,6,3,respectively.Rad-score,PV and PSAD showed statistical significance in univariate analysis.The odds ratio（OR）of t PSA failed to reach statistical significance.An AUC of 0.969 was achieved in the training cohort when merely using the Rad-score to diagnose PCa;The AUC in the validation cohort was 0.854.The diagnostic model of PCa consisted of Rad-score and PV,the equation of the model was: logit（P）=-2.186-0.045×PV+9.718×Rad-score.When used in the training cohort,the model obtained an AUC of 0.973,which was higher than that of the PI-RADS v2.1 score（AUC 0.838）.The diagnostic model achieved an AUC of 0.888 in the validation cohort.Conclusion: A greater PV is the protective factor for PCa.The diagnostic value of t PSA for PCa is limited.Overfitting has been observed in the training cohort.The diagnostic performance in the validation cohort can be improved when PV is added to Rad-score.Chapter Three Quantitative radiomics features and ADC values derived from multiple b-value diffusion weighted imaging: are they correlated with the pathological grade group of prostate cancer?Objective: To investigate the correlation between quantitative radiomics features,ADC value calculated from multiple b-value DWI,PI-RADS v2.1 score and the PCa grade group according to International Society of Urological Pathology（ISUP）.Materials and Methods:Patients undergoing prostate mp MRI in 6 medical centres were retrospectively enrolled.The b-values used in DWI were 50,400,800,1400s/mm².Pathologically-proven PCa lesions were included in this study,and the lesions were allocated to the corresponding ISUP grade group according to their Gleason scores.The VOI on T2 WI,DWI（b=1400s/mm²）,ADC map and DCE-MRI were manually delineated.Radiomics features were extracted after VOI delineation.The feature selection process was based on L1 regularisation and ICC analysis.ADC_50-400,ADC_400-800,ADC_800-1400,ADC_50-1400,ADC_{50-400-800-1400} and fh were calculated based on multiple b-value DWI using monoexponential model.Spearman correlation coefficients of the quantitative variables and the ISUP grade group were calculated.Either analysis of variance（ANOVA）or Kruskal-Wallis test was used to compare the quantitative variables in each subgroup.And boxplots for the statistically significant variables were plotted.Correlation between PI-RADS v2.1 score and ISUP grade group was assessed by calculating the Gamma coefficient.Z test was utilised to evaluate the statistical significance of the calculated Gamma coefficient.Results: 85 PCa lesions were included in the study.14 lesions were classified as ISUP grade group 1,21 lesions were ISUP grade group 2,21 lesions were ISUP grade group 3,11 lesions were ISUP grade group 4,18 lesions were ISUP grade group 5.ADC_50-400,ADC_400-800,ADC_800-1400,ADC_50-1400,ADC_{50-400-800-1400},fh and ADC measured were all significantly correlated with ISUP grade group,and the differences between subgroups were statistically significant.In terms of all the calculated ADC values,ADC_50-1400 had the highest spearman correlation coefficient（rho=-0.3789）.4 radiomics features were proven to be significantly correlated with ISUP grade group after calculating rho and comparing subgroups,including exponential＿glszm＿Size Zone Non Uniformity＿DWI,log-sigma-2-0-mm-3D＿glszm＿Low Gray Level Zone Emphasis＿DWI,wavelet-HLL＿glcm＿Idn＿ADC,wavelet-LHL＿glcm＿Idn＿DWI.Among the 4 radiomics features,log-sigma-2-0-mm-3D＿glszm＿Low Gray Level Zone Emphasis＿DWI was inversely correlated with ISUP grade group,whereas the other 3 radiomics features demonstrated positive correlation with ISUP grade group.The Gamma coefficient of PI-RADS v2.1 score and ISUP grade group was 0.358.The result of Z test showed that the calculated Gamma coefficient did not reach statistical significance.Conclusion: The ADC values and fh calculated from multiple b-value DWI using monoexponential model,along with the second-order and higher-order radiomics features extracted from high b-value DWI/ADC map,are significantly correlated with the ISUP grade group of PCa lesions.The correlation between PI-RADS v2.1 score and ISUP grade group is unremarkable.