Evaluation Of MRI-ADC Image Texture Analysis In Diagnosis Efficacy And Pathological Correlation Of Prostate Cancer

Objective:To investigate the value of prostate ADC images texture analysis in differentiating prostate cancer(PCa)from benign prostatic hyperplasia(BPH),and the correlation between ADC image texture characteristics and pathological grade of prostate cancer by extracting ADC image texture characteristics calculated from axial diffusion-weighted images at different b values.Methods:The image data of Mp-MRI(Multiparametric Prostate MRI)and biopsy pathology of relevant tissue performed for prostate-related symptoms in the Maanshan People's Hospital from January 2014 to July 2019 were collected.80 cases of prostate cancer and 54 cases of benign prostatic hyperplasia were included,with 134 cases in total.Two imaging diagnosticians working in the Department of Imaging for more than 5 years discussed and confirmed the area where the lesion was located on the axial DWI images using Fire Voxel software,delineated the regions of interest of cancer foci and proliferative foci layer by layer.The regions of interest in each layer composed the three-dimensional volume of interest(VOI)of the entire lesion,and the VOI was calculated and processed by the software to obtain the ADC images of tumor.The entropy,inhomogeneity,kurtosis,skewness,mean,and median of the ADC images within the VOI when the b value was taken as 400s/mm~2 and 1000s/mm~2,respectively,were calculated.The independent sample t-test was applied to compare whether the overall means of ADC image texture feature parameters were different between PCa group and BPH group with the b value of 400 s/mm~2 or 1000 s/mm~2.Receiver operator characteristic(ROC)analysis was applied to obtain the area under the curve(AUC),and the diagnostic efficacy of ADC image texture feature parameters in differentiating PCa from BPH and calculating the threshold for diagnosing prostate cancer at b values of400s/mm~2 and 1000 s/mm~2 were compared.All subsequent studies used b-values with high diagnostic power.Prostate cancer patients were divided into three groups according to Gleason score:highly,moderately,and poorly differentiated.One-way analysis of variance was used and then LSD was used for pairwise comparisons for the statistical differences of each texture feature parameter between the highly,moderately,and poorly differentiated groups.Independent sample t-test was used to compare whether the overall means of each ADC image texture feature parameter were different between the moderately and poorly differentiated group(Gleason score?7)and the highly differentiated group(Gleason score?6).The image texture characteristics with AUC> 0.7 were selected to establish a logistic regression model for multivariate-combined prediction of PCa and BPH.The diagnostic efficacy of ADC combined image texture characteristics parameters to differentiate PCa from BPH was evaluated according to the area under the ROC curve(AUC),and the threshold for the diagnosis of prostate cancer was calculated according to the ROC curve.Spearman correlation analysis was used to explore the correlation between each texture characteristics parameter and Gleason score.Results:1.When b value was 400 s/mm~2 and 1000s/mm~2,there were statistical differences in all ADC image texture characteristics between prostatic cancer group and benign prostatic hyperplasia group,and when b value was 1000 s/mm~2,there were significant differences in all texture characteristic parameters between the two groups(P<0.01).2.The AUC value of ROC curve analysis of each texture characteristics parameter,namely the diagnostic efficacy,was higher when b value was taken as 1000s/mm~2 than that when b value was taken as 400 s/mm~2.With a b value of 1000s/mm~2,the combination of entropy,mean,median and inhomogeneity had the highest power for the diagnosis of prostate cancer,with an AUC of 0.957.3.Entropy,mean and median are statistically different between moderately differentiated group and highly differentiated group.Entropy,mean,median,skewness,and kurtosis were statistically different between the poorly differentiated group and highly-differentiated group.Skewness and kurtosis were statistically different between the poorly differentiated group and moderately differentiated group.4.There was a correlation between Gleason score and entropy,mean,median,skewness,and kurtosis,with the greatest correlation for mean(r=-0.443,P<0.01),followed by kurtosis as well as median(r=0.437,-0.437,respectively,P<0.01).Conclusion:This study shows that texture analysis based parameter models established using magnetic resonance ADC images can be used for the differential diagnosis of prostate cancer and benign prostatic hyperplasia.The parameters extracted by texture analysis(entropy,skewness,kurtosis,inhomogeneity,mean,median)can provide new reference indicators for the assessment of prostate cancer pathological grade as well as invasiveness.