Research On The Predictive Value Of Magnetic Resonance Imaging Radiomics In Different Risk Stratification Of Prostate Cancer

Prostate cancer is a highly heterogeneous malignancy,early diagnosis and accurate evaluation of prostate cancer are directly related to patients’ survival prognosis and quality of life.Magnetic resonance imaging(MRI)has become one of the most important imaging examinations for the diagnosis of prostate cancer which can predict factors that represent invasiveness including pathological Gleason scores.MRI is critical to guiding clinicians in diagnosis and clinical decision-making.With the increasing amount of image data,how to use medical image data more effectively and extract more valuable information from tumor phenotype has become a research hotspot.In recent years depending on radiomics technology,medical images can be converted into excavated data using computer software and quantitative radiomics features can be extracted from mass data for the analysis of clinical information.The Prostate Imaging Reporting and Data System(PI-RADS)grade 1-5 for suspicious lesions of the prostate on MRI.The PI-RADS 3 lesions belong to benign or malignant equivocal lesions.It is defined as a“gray zone” scoring state between cancer and BPH.Especially when the PSA level of PI-RADS 3 patients is also in the“ gray area”of 4-10 ng/ml,it will be more difficult to differentiate the nature of the disease before operation.In this study,we analyzed the imaging features of patients with PI-RADS and PSA in the“double gray area” and their efficacy in differentiating benign from malignant disease using radiomics techniques.The AUC values of the predictive models can reach0.891 in the training group and 0.933 in the validation group respectively.Thus can provide clinicians a more accurate quantitative tool than the PI-RADS for clinical decision making.With the deep understanding of disease,the goal of the diagnosis of prostate cancer gradually changed from the detection of tumor alone to the detection of more clinically significant prostate cancer(Gleason score ≥7).Especially how to effectively screen the patients with clinically significant prostate cancer among the population of prostate specific antigen(PSA)level ≤20 ng/ml and to achieve accurate diagnosis and treatment had been the focus of clinical research.The purpose of this study was to evaluate the efficacy of MRI radiomics features in distinguishing between clinically significant and insignificant prostate cancer in patients at low and moderate PSA levels.The AUC values of the model can reach 0.863 in the training set and 0.839 in the validation set respectively.Thus this provide a favorable reference tool for the accurate diagnosis and treatment of relatively low-risk stratified population and help for clinicians in making differentiate decision.Perineural invasion(PNI)is a histological structure in which cancer cells surround and invade the peripheral nerve in the tumor microenvironment.Status of PNI positive have a higher incidence of tumor extracapsular invasion and is independently associated with poorer survival outcomes.However,there is no effective method to detect PNI in prostate cancer accurately by non-invasive methods,and the related research is still relatively blank.In this study,the effectiveness of magnetic resonance imaging features in predicting the status of PNI in high-grade prostate cancer was investigated by using radiomics techniques,and a combined model with high predictive performance was established that the AUC values of the model can reach 0.906 in the training group and0.947 in the validation group respectively.Thus this can provide a predictive method for preoperative detection of PNI status within high-grade prostate cancer and help urologists to make clinical decisions.Part Ⅰ Efficiency of MRI Radiomics Model on Differential Diagnosis of Prostate Cancer and Benign Prostatic Hyperplasia Among Population of PI-RADS 3 and PSA Level 4-10ng/mlObjective:To construct a radiomics model based on bi-parameter magnetic resonance imaging(MRI)features of patients with PI-RADS score of 3 and PSA level 4-10 ng/ml,and to explore the efficacy of the model in identifying prostate cancer and BPH in this population.Methods:The clinical data of patients with PI-RADS score of 3 were collected retrospectively.In all enrolled patients,the PSA levels were in the range of 4-10 ng/ml before MRI examination,prostate biopsy,transurethral resection of the prostate and radical prostatectomy.Primary imaging data including T2-weighted imaging(T2WI)and diffusion-weighted imaging(DWI)were collected.A total of the 106 patients were included in the study,38 had prostate cancer and 68 had benign prostatic hyperplasia disease.Regions of interest(ROI)in magnetic resonance images were manually delineated using ITK-SNAP software,and radiomics features were extracted from magnetic resonance images using Fe Ature Explorer(FAE)software.Multivariate logistic regression analysis was used to develop clinical feature models and radiologic feature models to predict prostate cancer versus BPH in patients with PI-RADS and PSA both in “gray zone”.Predictive performance was demonstrated using receiver operating characteristic(ROC)curves,calibration curves,and decision curve calculations.Results:The differential diagnostic efficiency of the clinical model was lower than that of the combined radiomics model(AUC = 0.866 and 0.933,respectively).In the training group,the combined imaging radiomics model(training group AUC: 0.891)showed better diagnostic efficiency than each individual sequence image model(T2WI training group AUC: 0.859,DWI training group AUC: 0.824).In the validation group,the combined imaging radiomics model(test group AUC: 0.933)showed better diagnostic efficiency than the T2 WI sequence image model(T2WI test group AUC: 0.855),lower diagnostic efficiency than the DWI sequence image model(DWI test group AUC: 0.945)but higher sensitivity and specificity.The comprehensive result show that the combined radiomics prediction model has the best differential diagnosis performance.Conclusion:In the population of PI-RADS score 3 and PSA level 4-10 ng/ml,both models based on clinical factors and based on radiomics characteristics were able to differentiate prostate cancer from BPH;However,the combined radiomics model is more effective than the predictive model including clinical factors,and is more helpful for differential diagnosis and individualized treatment.Part Ⅱ Efficiency of MRI Radiomics Model on Differential Diagnosis of Clinical Significant and Insignificant Prostate Cancer Among Low to Moderate PSA Level PopulationObjective:To construct a radiomics model based on bi-parametric magnetic resonance imaging features of prostate cancer,and to evaluate the value of the model in predicting clinically significant and insignificant prostate cancer among population of low to moderate PSA levels.Methods:Patients with prostate cancer confirmed by biopsy or radical prostatectomy pathology were retrospectively collected,and all patients had PSA levels in the range of <20 ng/ml before MRI examination.All patients underwent MRI before the invasive procedure.The original imaging data including T2-weighted imaging(T2WI)and diffusion-weighted imaging(DWI),as well as the clinical and pathological data of the patients were collected.A total of 82 patients were enrolled,including 45 patients with clinically significant prostate cancer at low and moderate PSA levels,and 37 patients with clinically insignificant prostate cancer at low and moderate PSA levels.Regions of interest(ROI)in magnetic resonance images were manually delineated using ITK-SNAP software,and radiomics features were extracted from magnetic resonance images using Fe Ature Explorer(FAE)software.A multivariable feature selection algorithm based on the least absolute shrink age and selection operator(LASSO)was used to filter the mass feature twice,and a multivariate logistic regression model was built.Predictive performance was assessed using receiver operating characteristic curve(ROC)analysis,and the area under the ROC curve(AUC)was calculated for quantification.The accuracy,sensitivity,specificity and predictive value were calculated,and the results of the logistic regression model were visualized by the predictive nomogram.Results:The clinical factors of this study sample were not statistically different between clinically significant prostate cancer and insignificant prostate cancer with low and moderate PSA levels.The combined model of bi-parameter magnetic resonance imaging radiomics features showed better diagnostic efficiency than each individual sequence image model in both the training group and the test group(training group AUC: 0.863,test group AUC: 0.839)(T2WI training group AUC: 0.732,test group AUC: 0.765;DWI training Group AUC: 0.809,test group AUC: 0.721).The best radiomics features can be extracted by stepwise regression to building final model,which was more efficient than the single sequence model.Conclusion:Bi-parameter MRI imaging model can accurately differentiate clinically significant prostate cancer from clinically insignificant prostate cancer in the population with low-to-moderate PSA levels.It is more helpful for individual prediction and treatment decision making using combine radiomics feature model than the single sequence model.Part Ⅲ Efficiency of MRI Radiomics Model on Predicting Peripheral Nerve Invasion Status in High-grade Prostate CancerObjective:To explore the model combined radiomics features based on magnetic resonance imaging(MRI)and clinical features in predicting peripheral nerve invasion status in high-grade prostate cancer.Methods:Patients with high-grade prostate cancer confirmed by biopsy or radical prostatectomy were retrospectively collected.All patients underwent MRI before the invasive procedure.The original imaging data including T2 weighted imaging(T2WI)and diffusion weighted imaging(DWI),as well as the clinical and pathological data of the patients were collected.Totally 183 patients were joined in group,54 had peripheral nerve invasion and 129 had no peripheral nerve invasion.Regions of interest(ROI)in magnetic resonance images were manually delineated using ITK-SNAP software,and radiographic features were extracted using Fe Ature Explorer(FAE)software.Using multivariable logistic regression analysis,three models were developed to predict peripheral nerve invasion in high-grade pathological grade prostate cancer: the clinical model,the MRI radiologic model,and a combined clinical-imaging model.Receiver operating characteristic curve(ROC),calibration curve,and decision curve calculations were used to demonstrate predictive performance.Results:The differential diagnostic efficiency of the clinical model was not statistically different from that of the radiologic model,with the AUC(area under curve)value of0.766 and 0.823,respectively.The combined radiomics model showed better diagnostic efficiency than each individual sequence image model in both the training group and the test group(training group AUC: 0.879,test group AUC: 0.908)(T2WI training group AUC:0.813,test group AUC: 0.827;DWI training group AUC: 0.749,test group AUC: 0.734).When radiomics and clinical models were combined,the identification efficiency of individual models were improved(training group AUC: 0.906,test group AUC: 0.947).Conclusion:Two-parameter MRI imaging radiomics model can accurately predict peripheral nerve invasion in high-grade prostate cancer.The predictive model combined with radiomics features and clinical factors may be helpful for individualized treatment.Conclusion1.Quantitative clinical factors model and MRI radiomics models allow for more accurate discrimination between benign and malignant lesions in patients with equivocal PI-RADS and PSA status compared with semiquantitative PI-RADS grading.MRI radiomics model is more effective by contrast.2.For prostate cancer populations with low and moderate PSA levels,there is a lack of sensitive and specific clinical factors to differentiate between significant and non-significant prostate cancer.MRI radiomics models can well identify low-risk and moderate-risk prostate cancers in this population to help differentiate clinical decision making.3.Bi-parameter MRI radiomics model can predict whether tumor cells invade peripheral nerves in pathological high-grade prostate cancer.Predictive model combined radiomics and clinical features can be helpful to predict the invasiveness of this tumor.4.Magnetic resonance imaging radiomics has good predictive value in prostate cancer risk stratification and can be used as a tool to predict disease related properties in prostate cancer patients at different risk and to help with accurate diagnosis and treatment.