Application Of Radiomics And Data Analysis By Magnetic Resonance Imaging In Detection,Risk Stratification And Nerve Invasion Of Prostate Cancer

Prostate cancer is a global public health problem that threatens human life and health,causing great harm to the male genitourinary system.According to statistics from the National Cancer Institute in 2021,Prostate cancer has the highest incidence of male tumors in 112 countries,ranking fourth in the world in new cancer cases and the fifth leading cause of male death.The mortality rate accounts for more than 97% of neoplastic diseases of the male reproductive system.The incidence of prostate cancer in China is not as high as that in Europe and the United States,but With the great changes in dietary structure and living environment,the aging population is increasing,and the incidence of prostate cancer is increasing year by year.Mp MRI is the best imaging method for the diagnosis of prostate diseases.It has important value in early diagnosis,treatment options,evaluation of efficacy and prognosis of prostate cancer.With the wide application of multi-parameter magnetic resonance imaging,more and more focal prostate cancer are detected and located accurately,which provides favorable conditions for the accurate treatment of early prostate cancer.In order to standardize the collection,interpretation and reporting of prostate magnetic resonance imaging,the ESUR developed the PI-RADS in 2012.ESUR in partnership with the American Academy of Radiology and the Ad Me Tech Foundation,released an updated version of PI-RADSv2,in 2015 and PI-RADSv2.1 in 2019.All PI-RADS versions provide guidance for image capture protocols and specifications,and the scoring system helps in the identification of biopsies and the management of clinically significant prostate cancer.In recent years,PI-RADS have been proposed,and artificial intelligence based on big data has developed rapidly,MRI-based imaging of the prostate has been widely carried out.Radiomics is the conversion of traditional medical imaging images into data that can be mined and high-throughput quantitative analysis.It breaks through the traditional medical imaging model based on morphology and semi-quantitative analysis,and integrates digital imaging information,Methods such as statistics and machine learning can fully dig and analyze the additional information hidden in the image,and make the most efficient use of the results of imaging examinations.Radiomics can show the heterogeneity of tumors more comprehensively and quantitatively than morphological visual analysis.It can be used in a variety of medical images to assist clinical diagnosis and select treatment,and supplement the traditional diagnostic methods.The analysis method of radiomics provides a noninvasive tool for evaluating the biological characteristics and heterogeneity of prostate cancer.The current applications of radiomics in prostate cancer include tumor diagnosis and staging,curative effect evaluation and prognostic analysis,and gene prediction.Based on this background,we conducted this study.This study consists of three chapters: the first chapter is the detection of clinically significant prostate cancer by texture analysis based on biparametric magnetic resonance histogram;the second chapter is the Risk stratification of prostate cancer based on biparametric magnetic resonance imaging radiomics machine learning;the third chapter is the evaluation of the value of magnetic resonance radiomics nomogram in predicting perineural invasion of prostate cancer before operation.Chapter Ⅰ: Detection of clinically significant prostate cancer by texture analysis based on biparametric MRI histogramBackground and objective: The incidence of prostate cancer(PCa)ranks first among the most common malignant tumors in elderly men in the world.At present,with the continuous improvement of the economic situation of Chinese residents and changes in their way of life,the incidence of prostate cancer in China is showing a growing trend,which has become a major health problem affecting the normal life of many families.The European Urological Association’s PCa guidelines in 2017 recommend active monitoring and followup of PCa patients with a Gleason score(GS)< 7,while clinically significant prostate cancer(CSPC)patients with GS ≥ 7 should take timely treatment and intervention because of their increased risk of progression and short overall survival.Therefore,it is very important to select the best treatment for these patients through accurate risk assessment.Radiomics texture analysis can evaluate the spatial relationship of pixel intensity and be used to quantify the heterogeneity of lesions.Extracting quantitative data invisible to the naked eye from standard medical images can make diagnosis more accurate and help optimize clinical decision-making.In this study,the texture features of biparametric MRI of prostate nodules and the whole prostate were extracted,and texture analysis was used to detect clinically significant prostate cancer,and to evaluate the efficacy of PI-RADS,clinical index,texture and the combination of the three in the detection of clinically significant prostate cancer.Methods: The clinical data of 260 patients with prostatic nodules were collected.They all had definite pathological results.All patients with non-cancerous prostatic nodules underwent targeted biopsies under the guidance of ultrasound,and all patients with cancerous lesions underwent robotic radical prostatectomy and total resection(RP).All the pathological specimens were analyzed by HE staining and immunohistochemistry.Among them,106 cases had clinically significant carcinoma(Gleason score ≥7)and 154 cases had non-clinically significant carcinoma(Gleason score < 7).All patients underwent 3.0TMRI with the same sampling parameters,and the clinical risk factors(age,lesion size,prostate volume,PSAD,PSA)and PI-RADS score related to prostate cancer were analyzed.The research is based on the biparametric MRI of the prostate sequence(including transverse T2 WI and ADC images),using the prototype multiparameter magnetic resonance analysis software Multiparametric Analysis v1.2.0(Siemens Healthineers,Erlangen,Germany)is used for image calibration,focus segmentation and extracting histogram texture features for analysis.Chi-square test and analysis of variance were used to compare the differences in clinical characteristics.Twenty-two texture parameters were analyzed by univariate method.When they were normally distributed,the parameters were statistically compared by doublesample t test.In the case of non-normal distribution,statistical comparison was made by Mann-Whitney U test.Univariate analysis was used to identify the significant variables in the texture,and the AUC value of each variable in the detection of clinically significant cancer was calculated,and the efficacy was compared.Multiple logistic regression was used to establish the discriminant model,and the receiver operating characteristic(ROC)curves of each model were generated.The area under the curve(AUC),accuracy,sensitivity and the specificity of the optimal cut-off point were calculated.Delong test was used to compare the differences of ROC curves of different models.Result:The models were established based on the whole lesion and the whole gland respectively.When texture analysis was performed to detect clinically significant cancer in transitional zone and peripheral zone,the AUC of texture model was 0.903(95% IC: 0.853-0.953)and 0.859(95%IC:0.764-0.954),and the AUC of texture-clinical model was 0.938(95%IC:0.819-0.974)and 0.894(95%IC:0.819-0.968),respectively.The AUC of the combined PI-RADS model is 0.971(95%IC:0.951-0.991)and 0.888(95%IC:0.818-0.957),respectively.The texture analysis of the whole gland showed that the AUC of the texture model was 0.850(95%IC:0.787-0.913)and 0.672(95%IC:0.555-0.789),and the texture-clinical model AUC was 0.919(95%IC:0.642-0.965)and 0.751(95%IC:0.6420.861),respectively.The AUC of the combined PI-RADS model was 0.981(95%IC:0.965-0.996)and 0.804(95%IC:0.703-0.905),respectively.No matter the texture analysis of whole focus or whole gland,texture model,texture-clinical model and combined PI-RADS model have high detection efficiency for clinically significant cancer of transitional zone,and texture model,texture-clinical model and combined PI-RADS model have higher predictive ability for clinically significant cancer of peripheral zone in whole-focus texture analysis.Conclusion: The texture analysis based on the histogram of the biparameter MR of the whole lesion and the whole gland can detect clinically significant prostate cancer,and can improve the diagnostic efficiency of PI-RADS for clinically significant prostate cancer,especially when combined with texture analysis and clinical factors.Chapter Ⅱ Risk stratification of prostate cancer based on biparametric magnetic resonance imaging radiomics machine learningBackground and objective: prostate cancer(PCa)is a global public problem that threatens human health and life.its mortality accounts for more than 97% of male reproductive organ neoplastic diseases.The incidence of prostate cancer is increasing.Gleason grade of prostate cancer is one of the most powerful predictors of progression and survival of prostate cancer,and it is also a decisive factor affecting the treatment of PCa.Gleason score(GS)system identifies the prostate histologically by analyzing the degree of gland differentiation,and reflects the heterogeneity of the tumor according to the primary score and secondary score.According to GS,PCa can be divided into two groups: low-risk PCa(GS ≤ 3 + 4)and high-risk PCa,(GS ≥ 4 + 3).Patients with GS≤3+3 have similar prognosis to patients with GS=3+4,,and is significantly better than GS ≥ 4 + 3 patients.Imaging has been widely used in tumor research in recent years.The great advantage of imaging science is that it can automatically filter the comprehensive data extracted from the image,and can identify the heterogeneity of tumors from macroscopic images and molecular changes of genes or proteins.With the characteristics of non-invasive,rapid,repeatable and low cost,it has been widely used to predict the prognosis of head and neck,colorectal cancer and lung cancer.The purpose of this study is to extract the biparametric MRI features of prostate cancer,construct a model and evaluate the imaging ability to distinguish high-grade and low-grade prostate cancer,stratify the risk of prostate cancer,and help optimize the treatment plan.Methods: The clinical data of 128 patients with prostate cancer were collected.They all had definite pathology and underwent robot-assisted radical prostatectomy.,and large pathological sections were obtained.Among them,60 cases were in the low-risk group(Gleason score ≤3+4)and 68 cases were in the high-risk group(Gleason score ≥4+3).All patients underwent 3.0TMRI with the same sampling parameters,and clinical risk factors associated with prostate cancer(age,lesion location,lesion volume,PSA and PI-RADS scores)were analyzed.This study was based on biparametric MRI(Bp MRI)sequences(including transverse T2 WI and ADC images).After image preprocessing,lesions were segmented and radiomics features were extracted for analysis.The minimum absolute contraction and selection operator algorithm(LASSO)was used to extract the image omics features from the training data set.Patients were randomized into training set and validation set in a 7:3 ratio,and pathological risk stratification for prostate cancer was performed.Determined by multiple logistic regression analysis to prostate cancer risk stratification of independent risk factors,to establish the forecast model of the prostate cancer risk stratification: clinical model,radiomics model and the clinical-radiomics combined model,and build radiomics nomogram,nomogram to evaluate performance in the training data set,and is verified in the validation data set.Receiver operating characteristic(ROC)curve analysis,decision curve and calibration curve were used to compare the diagnostic performance,clinical benefit and pathological coincidence rate of each model.Results: PSA and PI-RADS scores could be used as predictors of risk stratification(P < 0.05).For risk stratification of prostate cancer,the predictive effect of the imaging model and the clinical-radiomics combined model were the same [AUC: 0.78(95%CI: 0.63-0.93)] and both were better than that of the clinical model [AUC: 0.75(95%CI: 0.60-0.91)].Decision curve analysis showed that the imaging omics model and the clinical-radiomics combined model had higher clinical net benefits than the clinical model.The calibration curve indicated that the prediction results were in good agreement with the pathological results.Conclusion: Biparametric MRI radiomics model can effectively stratify prostate cancer risk,and clinical-radiomics machine learning model based on biparametric MRI radiomics can improve the accuracy of risk stratification prediction for prostate cancer compared to evaluating only clinical risk factors associated with prostate cancerChapter Ⅲ Evaluation of the value of MRI radiomics nomogram in predicting Perineural invasion of Prostate Cancer before OperationBackground and purpose: The feature analysis of radiomics can be used as biological markers for disease prediction and prognosis.The degree of resection of neurovascular bundles in radical prostatectomy(RP)is closely related to postoperative function(urinary control or erection)and recurrence-free survival.Perineural invasion(PNI)is an invasive tissue disease of PCa Physiological characteristics are also an important indicator of its prognosis.Judging the patient’s PNI condition before RP surgery is very important for choosing the best option for surgery and patient prognosis.This study aims to construct and verify a combined clinical-radiomics model(nomogram)based on MRI,including radiomics features,clinical factors and related features,to predict the preoperative PNI status of PCa patients.Methods: 338 patients with prostate cancer confirmed by surgery and pathology after robot-assisted radical prostatectomy in our hospital from March 2015 to September 2020 were analyzed retrospectively.The PNI status of the patients was recorded(212 positive and 126 negative).The images of 246 patients were collected by GE750 MR and 92 patients were scanned by Siemens Skyra MR.The imaging features extracted from T1 WI,T2WI and DWI were used to establish the imaging model.Collect the clinical characteristics and pathological variables of patients and establish a clinical model,and finally further integrate the radiomics model with the clinical model to establish a clinical-radiomics combined model(nomogram).According to the proportion of 7:3,the patients scanned by GE750 MR were randomly divided into training group(n = 174)and internal verification group(n = 72),and 92 patients scanned by Siemens Skyra MR were taken as external verification group.The minimum absolute contraction and selection operator regression(LASSO)algorithm were used to establish the radiomics model.The diagnostic performance of different models was evaluated by calculating the area under the curve(AUC),and Delong test was used to compare the differences of AUC between models.Calibration curve and decision curve analysis are used to evaluate the calibration and clinical practicability of models.Results: The AUC values of the radiomics model in the training group and the internal validation group were 0.82 and 0.60,respectively.The AUC values of the clinical model in the training group and the internal validation group were 0.75 and 0.71,respectively.The AUC values of clinical-radiomics combined model(nomogram)in training group and internal verification group were 0.84 and 0.66,respectively.Decision curve analysis(DCA)shows the clinical utility of the clinical model and the combined model of clinical-radiomics(nomogram).The accuracy,sensitivity and specificity of the radiomics model in the internal validation group were 59.7%,62.5% and 54.2%,respectively.The accuracy,sensitivity,and specificity of the clinical model in the internal validation group were 70.8%,77.1%,and 58.3%,respectively.The accuracy,sensitivity,and specificity of the combined clinicalradiomics(nomogram)in the internal validation group were 63.9%,76.2%,and 46.7%,respectively.In external verification,the clinical model is better than the combined model(AUC: 0.82 vs.0.69,P <0.001),and the accuracy,sensitivity,and specificity are 77.2%,98.0%,and 53.5%,respectively.Conclusion: the combined clinical-radiomics model(nomogram)based on magnetic resonance imaging can distinguish between PNI and non-PNI lesions,which is helpful to predict the state of PNI before prostatectomy,which is not superior to the clinical model.PI-RADS score and clinical T stage are independent predictors of PNI.