Study On The Prediction Of Bone Metastasis Of Prostate Cancer By Ultrasound Radiomics

Objective To establish prediction model for bone metastasis of prostate cancer based on the radiomics features of prostate cancer combined with clinical indicators to evaluate the diagnostic value of the model in this study.Method A total of 123 patients with prostate cancer were included in the study,and they were divided into prostate cancer BM group(46 cases)and prostate cancer NBM group(77 cases)according to the results of prostate biopsy or surgical pathology,SPECT / CT and bone related tests.All patients underwent transrectal contrast-enhanced ultrasound,transrectal ultrasound-guided prostate biopsy and MRI examination.Grayscale features of contrast-enhanced ultrasound images of prostate cancer patients were extracted by Traite Ima2 DTime Various.exe self-compiled software,and contrast-enhanced parameters,contrast enhancement patterns and ADC values of MRI were recorded.The clinical indicators of patients were collected,including serum PSA,serum ALP,serumcalcium,serum phosphorus and Gleason score.The decision tree chi-square automatic interactive detection(CHAID)algorithm was used to establish the prediction model of prostate cancer BM,and the cross validation method was used for internal verification to calculate the risk and accuracy of prediction.Result(1)A total of 123 prostate cancer patients were included in this study,including 46 patients with BM(37.4 %)and 77 patients with NBM(62.6 %).Of the46 BM patients,30(65.2 %)were multiple BM and 16(34.8 %)were single BM.Among them,there were 28 cases of pelvic metastasis(60.9 %),28 cases of thoracic metastasis(60.9 %),27 cases of vertebral metastasis(58.7 %),17 cases of limb metastasis(40.1 %),and 6 cases of skull metastasis(13.1 %).(2)The results of single factor analysis showed that the serum t PSA,serum f PSA,f PSA / t PSA,serum ALP and serum P in the BM group of prostate cancer were higher than those in the NBM group(P < 0.05).With the increase of Gleason grading,the proportion of BM patients increased(P = 0.008),and 65.2%(30 / 46)of BM patients were concentrated in grade 4-5,while 61.1 %(47/ 77)of NBM patients were concentrated in grade 1-3.(3)There was no difference in prostate volume,contrast-enhanced ultrasound mode between BM group and NBM group(P > 0.05);There was significant difference in contrast-enhanced ultrasound parameters PI,MTT,AUC and HT between BM and NBM groups of prostate cancer(P < 0.05).(4)The gray scale images and gray scale characteristics of contrast-enhanced ultrasound between the two groups were compared.The differences in contrast-enhanced ultrasound and gray scale center distance from 1 to 8,average contrast-enhanced ultrasound and gray scale center distance,and average gray scale of contrast-enhanced ultrasound between the BM group and the NBM group significant(P < 0.05).(5)Compared with the ADC values in the NBM group and the BM group,the ADC values in the BM group were lower than those in the NBM group,and the difference was statistically significant(P < 0.05).(6)The decision tree model for predicting prostate cancer BM was constructed by single factor analysis of statistically significant characteristics and indicators.The results showed that the root node was the average center distance of contrast-enhanced ultrasound,and the sub-node was HT,PI and ALP.The overall prediction accuracy of this model for prostate cancer BM was 85.8 %,that of NBM group was 93.5 %,and that of BM was 69.9 %.The re-entry risk and cross-validation risk are 0.154 ± 0.033 and 0.236 ± 0.038,respectively.Conclusion The BM prediction model established by ultrasound radiomics features of prostate cancer combined with clinical indicator ALP can provide reference for personalized prediction of prostate cancer bone metastasis.