T2WI Texture Analysis Was Used To Evaluate The Efficacy Of Androgen Deprivation Therapy For Prostate Cancer

Objective:To evaluate the efficacy of androgen deprivation therapy for prostate cancer and the value of detecting prostate cancer after treatment by T2 WI texture analysis.Methods:Retrospective analysis of patients with prostate cancer confirmed by biopsy in the affiliated Hospital of Inner Mongolia Medical University from January 2017 to December 2018 after ADT treatment(the median treatment time was 7 months).Routine T1 WI,T2WI and DWI scanning were performed after treatment.Using ITK-SNAP software to manually sketch ROI.layer by layer on T2 WI.308 texture features are extracted by AK software of GE company.Intra-group correlation coefficient(ICC)was used to evaluate feature repeatability.Independent sample t-test or Mann-Whitney U test was used to screen out texture features with statistically significant differences between groups.Lasso regression model and 10-fold cross-validation method were used to further screen and model the features.Multi-factor logical regression model was finally constructed to build three machine learning models.ROC curve and decision curve(DCA)were used to evaluate the diagnostic effectiveness of the model.Results:After ADT treatment,23 patients with lesions were confirmed by puncture biopsy,20 patients without lesions,the age was(66.6 ±7.4)years old.Before treatment,the mean value of prostate specific antigen((prostate specific antigen,PSA)in "focus group" was60.58 ng / ml,and that in "No lesion group" was 77.52 ng / ml,P = 0.79,the difference was not statistically significant.Among the 308 features extracted,ICC is between 0.43 and 0.99.59 texture features with ICC ? 0.8 are deleted in model 1,36 texture features with ICC ? 0.8are deleted in model 2,and 54 texture features with ICC ? 0.8 are deleted in model 3.Among the texture features with good consistency,the texture features that have no statistical significance are deleted by t-test or Mann-Whitney U test,and the texture feature combinations with the highest diagnostic efficiency are screened out by Lasso regression model and 70% discount cross-verification,and the four features Correlation?angle135?offset4,Haralick Correlation?All Direction?offset4?SD,Elongationand Low Intensity Large Area Emphasis are screened out to construct model one.In the first model,the area under the ROC curve of prostate cancer after ADT treatment was 0.87,the sensitivity was 0.739,the specificity was 0.75,and the accuracy was 0.74.Two texture features Voxel Value Sum and Long Run Emphasis?angle45?offset1 were selected to construct model 2.The area under the ROC curve was 0.91,the sensitivity was 0.81,the specificity was1,and the accuracy was 0.89.The three features of GLCMEntropy?All Direction?offset7?SD,Long Run Emphasis?angle135?offset4,Long Run High Grey Level Emphasis?All Direction?offset4?SDNULLADC were selected to construct model 3.The area under ROC curve,sensitivity,specificity and accuracy of distinguishing cancer from peripheral zone were 0.87,0.952,0.67 and 0.81 respectively.In model 1,the DCA curve represents the net benefit value under different risk thresholds.When the risk threshold is 10%-99% in the training group,the net benefit value obtained by using texture analysis to evaluate the therapeutic effect of prostate cancer ADT is greater than the net income value of all intervention and no intervention.Conclusion: T2 WI texture features can evaluate the different efficacy of ADT treatment,and T2 WI texture features can also detect prostate cancer after ADT treatment.Early measurement of the response to treatment in vivo may be a powerful tool to predict the outcome and help to select patients who need further adjuvant therapy.As a predictive model,texture features are feasible in evaluating curative effect.In the later stage,by increasing the amount of data,it is expected to further increase the accuracy of model diagnosis.Texture features can be used as independent criteria for whether ADT patients need further EBRT intervention.In terms of tissue classification function,the T2 WI texture features of model 2and model 3 are better than the corresponding histogram parameters,which proves that T2 WI texture features can well distinguish the tumor from the surrounding benign tissue after ADT treatment.