Radiomics Research Of Transitional Prostate Cancer Based On MpMRI

BACKGROUD: Prostate cancer occurs predominantly in people over the age of 60,ranking second in the global male cancer incidence [1].In the European and American countries,the incidence of prostate cancer(prostate cancer,PCa)ranks first among male malignant tumors [2].Although the incidence of prostate cancer in China is much lower than that in European and American countries,it has been increasing year by year in recent years.Prostatic hyperplasia is a common disease of middle-aged and elderly men in China,and the incidence rate gradually increases with age.Prostate cancer mostly occurs in the peripheral zone of the prostate,and benign prostatic hyperplasia mainly occurs in the transition zone.However,a large number of clinical cases have found that prostate cancer that occurs in the transition zone is not uncommon,and sometimes it is difficult to distinguish between the two.The aggressiveness of prostate cancer is an important factor that affects the treatment and prognosis of patients.Medium and low-risk prostate cancer has a relatively low malignancy and relatively slow tumor growth.The prognosis and survival time of patients are relatively long.Prostate cancer can be observed and waited for with active detection.For high-risk prostate cancer,due to the relatively high degree of malignancy,the prognosis and survival time of patients are relatively short,and more aggressive treatment methods(radiochemotherapy or radical resection)are used clinically [3].Puncture biopsy is currently the "gold standard" for preoperative diagnosis of prostate cancer and assessment of prostate cancer aggressiveness.However,as a invasive examination,biopsy can cause various complications.Therefore,an accurate non-invasive diagnosis method is needed in clinic to diagnose prostate cancer and evaluate the invasiveness of prostate cancer,so as to reduce the harm to patients and improve patient compliance.Texture analysis(TA)is a new image evaluation technology.By analyzing the distribution of gray values and pixel values in the image,it can mine small structures that are difficult for human eyes to recognize in the image.Prognosis judgment and efficacy evaluation [4].Machine learning is a science about artificial intelligence.The main research in this field is artificial intelligence,which can use known data or previous experience to optimize the performance standards of computer programs [5].With the rise of artificial intelligence in recent years,more and more research is devoted to the application of artificial intelligence in medicine.A large number of studies have confirmed that texture analysis and machine learning models have great significance and extensive clinical application prospects in the qualitative,localization,grading,efficacy evaluation and prognosis of tumors.OBJECTIVE: This paper studies the feasibility of multi-parametric magnetic resonance imaging(mp MRI)texture analysis combined with machine learning model in the prediction of benign and malignant lesions in prostate transition zone and the invasion of prostate cancer.METHODS: Part 1: Retrospective collection of 100 patients each who underwent MRI examination in our hospital from January 2015 to December 2019 and were pathologically confirmed as prostate cancer and benign prostatic hyperplasia.Collect patient imaging data,use Ma Zda software to delineate the entire prostate tissue on the largest level of the lesion on T2 WI and ADC images,extract texture parameters to establish data sets,and establish T2 WI data sets,ADC data sets and total data sets(T2WI data,respectively)Set + ADC data set)and then use the Pearson + PCA method to perform feature dimensionality reduction on the data set respectively,and filter out the parameters that are meaningful for diagnosis.Establish a machine learning model: randomly select 70% cases in the prostate cancer group and BPH group as the training group,and the remaining 30% cases as the verification group.Five machine learning models were established in the training group,namely Decision Tree(DT),Naive Bayesian(NB),K-Nearest(KNN),Random forests(RF),Support vector machine(Support vector machine,SVM).Use the test group to verify the model and evaluate the effectiveness of each model.Part ?: We divided 100 cases in the prostate cancer group according to their pathological Gleason score(GS)classification into a low-risk group(GS?7)and a high-risk group(GS?8).Excluding 17 cases lacking GS scores,83 cases that meet the criteria were finally included,among which 30 were low-risk(GS?7)and 53 were high-risk(GS?8).Collect patient imaging data,use Ma Zda software to delineate the entire prostate tissue on the largest level of the lesion on T2 WI and ADC images,extract texture parameters to establish a data set,use Mazda's own dimensionality reduction method to reduce dimensionality,and pass the B11 module After classification,the misjudgment rate of different dimensionality reduction methods and different classification methods is obtained.The 30 texture parameters obtained by FPM dimensionality reduction were subjected to T test,and the statistical parameters with statistical significance(P <0.05)were selected.The intentional texture parameters were included in the Logis regression to obtain the discrimination of low and medium risk(GS?7)High-risk(GS?8)independent risk factors,and do ROC curve evaluation parameters to predict performance.RESULTS: The prediction performance of each model based on the T2 WI dataset is: DT(AUC = 0.65,accuracy rate = 76%),NB(AUC = 0.83,accuracy rate = 82%),KNN(AUC = 0.74,accuracy rate = 78%),RF(AUC = 0.81,accuracy rate = 83%),SVM(AUC = 0.65,accuracy rate = 69%);the prediction performance of each model based on the ADC dataset is: DT(AUC = 0.72,accuracy rate = 71%)NB(AUC = 0.90,accuracy rate = 88%),KNN(AUC = 0.73,accuracy rate = 71%),RF(AUC = 0.84,accuracy rate = 83%),SVM(AUC = 0.67,accuracy rate = 69%));The prediction performance of each model based on the total data set is: DT(AUC = 0.87,accuracy rate = 84%),NB(AUC = 0.93,accuracy rate = 90%),KNN(AUC = 0.84,accuracy rate = 88%)),RF(AUC = 0.84,accuracy rate = 83%),SVM(AUC = 0.80,accuracy rate = 78%).The NB model has better diagnostic performance than the other four models in the models based on the T2 WI data set,the ADC data set,and the total data set,achieving very good and excellent performance.The performance of each model based on the ADC dataset is better than the T2 WI dataset.In the three data sets,the NB model trained based on the total data set has the best performance compared to other models(AUC = 93%,accuracy rate= 90%).Based on the texture analysis of the invasiveness of transitional zone prostate cancer,it was found that compared with T2 WI,the ADC sequence contains more texture features that distinguish between low-risk and high-risk transitional zone prostate cancer.The independent influencing factors of the overall texture of the prostate to predict the invasion of prostate cancer in the transition zone are Wav En HH?s-5 and Wav En HL?s-5.Using Wav En HH?s-5 and Wav En HL?s-5 parameters for the diagnosis of prostate cancer,AUC were 0.799 and 0.765,respectively,showing good diagnostic performance.The combined diagnostic performance of the two parameters is better,with an AUC value of 0.835,showing very good diagnostic performance.CONCLUSION: The mp MRI prostate cancer imaging omics model can better diagnose transitional zone prostate cancer.Texture analysis is of great significance in predicting the invasiveness of transitional zone prostate cancer.Prostate cancer imaging omics has good clinical application potential and is expected to become a radiologist's auxiliary diagnostic tool in the future.