The Study On The Prediction Of Prostate Cancer By Ultrasonic Radiomics Combined With Apparent Diffusion Coefficient Of MRI

Objective:To explore the value of gray-scale ultrasound and contrast-enhanced ultrasound radiomics features combined with MRI apparent diffusion coefficient value in predicting prostate cancer.Methods:A total of 212 patients with clinically suspected prostate cancer were collected,118 cases of prostate cancer and 94 cases of benign prostatic lesions confirmed by pathology.The clinical indicators,ADC value of MRI,gray-scale ultrasound and contrast-enhanced ultrasound data were analyzed.The time-intensity curve was drawn by QLAB software.The rise time(RT),wash in slope(WIS),peak intensity(PI),time to peak(TTP),area under the curve(AUC),intensity drop half time(HT)and mean transit time(MTT)of the region of interest(ROI)of the prostate were quantitatively analyzed.The imaging video was intercepted into a static image,and the Image J software was opened to manually outline the ROI.The self-made image processing program was used to analyze the ROI and extract the gray feature.Including contrast-enhanced ultrasound and gray-scale ultrasound image extraction of gray-scale mean,standard deviation,center distance 1 to center distance 8 a total of 20 parameters,calculate the range of contrast-enhanced gray-scale features and the average gray-scale features of gray-scale ultrasound.The clinical indexes,ADC values,gray-scale characteristics of radiomics,and time-intensity curve radiomics parameters of contrast-enhanced ultrasound between prostate cancer group and benign prostate disease group were compared,and the gray-scale characteristics were clustered by principal component analysis.All patients were randomly divided into the modeling group(n=148)and the verification group(n=64)according to the ratio of 7:3.In the model group,the diagnostic model for predicting prostate cancer was established by Logistic regression analysis,and the data of the validation group were used for verification.Receiver operating characteristic curve(ROC)of modeling group and validation group for evaluating the diagnostic efficiency of prostate cancer.Hosmer-Lemeshow test and calibration curve were used to evaluate the calibration of the prediction model.Result:(1)The age,PSA value and ADC value of MRI in the prostate cancer group were higher than those in the benign prostatic lesion group,and the difference was statistically significant(P<0.05).(2)Patients in the prostate cancer group showed higher enhancement,rapid progression and rapid regression mode than those in the benign prostate disease group.Patients in the benign prostate disease group showed more homogeneous enhancement,while patients in the prostate cancer group showed more nodular or inhomogeneous enhancement.(4)The contrast-enhanced ultrasound radiomics parameters PI,AUC and WIS of prostate cancer group were higher than those of benign prostate disease group(P<0.001).The gray characteristics of 20 omics images extracted from gray-scale ultrasound and contrast-enhanced ultrasound images were analyzed by single factor analysis.The range of contrast-enhanced gray center distance from 1 to 8,the range of contrast-enhanced gray mean,and the mean of gray-scale gray standard deviation were statistically significant between groups(P<0.05).Principal component analysis showed that the cumulative contribution of four principal component feature roots representing the gray feature of radiomics was 80.755%,which could represent the main information of gray feature.(5)Age,PSA,ADC value,PI,AUC,WIS and four principal components of univariate analysis(P<0.05)were included in the Logistic regression.The results showed that age,PSA,ADC value,PI and the first principal component(FAC-1)were independent risk factors for prostate cancer(P<0.05).The diagnostic model Logistic(P)=-5.283+0.103×(age)+0.031×(PSA)-5.070×(ADC value)+0.329×(PI)+1.116×(FAC-1)was established.The area under the ROC curve was 0.925,the cut-off value of the diagnostic model was 0.76,the sensitivity was 72.5%,the specificity was98.5%,the Youden index was 71.4%,the positive predictive value was 95.6%,and the negative predictive value was 73.8%.Validation group of the area under the ROC curve was 0.922,the diagnostic threshold was 0.76,the sensitivity was 73.68%,the specificity was 92.31%,the Youden index was 66.0%,the positive predictive value was 93.3%,and the negative predictive value was 70.6%.Hosmer-Lemeshow test showed that there was no significant difference between the prediction probability and the actual probability of the modeling group and the verification group(χ~2=13.86,P=0.127 in the modeling group andχ~2=8.33 P=0.501 in the verification group).The calibration curve reflected the prediction probability and the actual probability had good consistency.Conclusion:The radiomics features extracted by contrast-enhanced ultrasound and gray-scale ultrasound are important in the differential diagnosis of benign and malignant prostate diseases.Prostate cancer tissues have lower ADC values than benign prostate tissues.The diagnostic model established by the combination of ultrasonic radiomics features,ADC values of MRI and clinical indicators can provide valuable information for the noninvasive prediction of prostate cancer.