Preliminary Study On The Value Of Texture Analysis Of MRI In Diagnosing Bone Metastases In Prostate Cancer

Objective: This paper intends to use mpMRI images for texture analysis,train machine learning models,and initially explore its diagnostic value in bone metastasis of prostate cancer.Methods: The study cohort consisted of pathologically confirmed prostate cancer patients who underwent magnetic resonance examination in our hospital from January 2016 to September 2019.112 patients were included in the study.Match the T2,ADC images and pathological results to determine the lesion boundary.ITK-SNAP software was used to manually layer the region of interest(ROI)to obtain the volume of interest(VOI)of the lesion.The A.K.software was used to extract the texture parameter features,select features and develop the predicting model,and train these model classifiers.Statistical analysis and ROC curve drawing were performed using MedCalc software(Version 3.1.0).R software version 3.6.2 was used to draw the decision curve to verify the clinical value of the model.Results: There are 5 ADC texture features and 7 T2 texture features,which can be used to predict simultaneous bone metastases in the prostate cancer.Among the six machine learning models,the area under the ROC curve(AUC)of the training group of the Logistic Regression classifier model based on ADC image is 0.767,and the test group has an AUC of 0.790,showing moderate classification performance.The AUC value of the C＿SVC & LINEAR classifier model based on the T2 image in training group is 0.760,and the AUC of the test group is 0.806,which shows better prediction ability.In the decision curves of the two models for predicting bone metastasis of prostate cancer,the threshold probability range of the C＿SVC & LINEAR classifier model based on T2 texture features is 0.2-0.8,with a large net benefit and high clinical value.The threshold probability range of the Logistic Regression classifier model based on ADC texture features is 0.2-0.4,which indicates that the clinical application value of this model to predict bone metastasis in prostate cancer is limited.Conclusions: The texture features of radiomics can be applied to the prediction of clinical bone metastasis in prostate cancer.Based on the texture features of MRI,machine learning model for predicting bone metastases of prostate cancer with good classification performance can be established.This has great potential and clinical value for the choice of treatment methods and prognosis evaluation of patients with prostate cancer.