CT Radiomics Model Research Of Renal Space-occupying Lesions

Renal space-occupying lesions,as common kidney diseases,its morbidity is increasing constantlyand.With the popularization of physical examination and the improvement of self-protection consciousness,the diagnostic rate of early RSOL and the detection rate of benign tumors are rising.However,there are still some difficulties in the differential diagnosis of patients just by the approach of observing imaging images visually,and such method may results in obscure accuracy of diagnosis,misdiagnosis and missed diagnosis.Therefore,how to make more accurately and effectively preoperative diagnosis and avoid unnecessary nephrectomy caused by misdiagnosisare particularly important,which also are meaningful and challenging tasks.Radiomics,as a study of interdiscipline with medical imaging,clinical manifestation,genes and other information,makes fully use of computer technology to analyse images.Recently,the application of radiomics in RSOL has gradually been widespread,radiomics can extract and analyse massive imaging features from medical images such as CT and MRI,then construct models based on those features,and offer the analysis and prognosis of diseases,which is overwhelmingly significant for the clinical assistance of accurately therapeutic stragey.To provide assistance for the clinical diagnoses,this paper studies the RSOL which is common and easy to misdiagnosed based on radiomics,The detailing contents of this paper are as following:1.Pretreatment of experimental data.CT images provided by Guangdong Southern Hospital,which belonged to confirming ailing patients,were attributed to sample set.The region of interest was delineated manually by clinicians,then the volume of interest was reconstructed by three-dimensional volume reconstruction.Be based on the volume of interest,362 features including geometric features,statistical features and texture features are extracted by MATLAB toolkit,to offer subsequent feature selection and classification model construction.2.Clinical application of radiomics in study whether hydronephrosis is associated with calculi associated with renal cell carcinoma.This study proposed the feature selection model(SVM-InfFS)to remove the unrelated features,redundant features and noise interference,meanwhile,Support Vector Machine was used to construct classification model;besides,research compared the effect on image classification among different feature sets and feature selection method,finally analysed and validated the feasibility whether the constructed classifier in this paper can be applied to clinical assistant diagnosis after combing with diagnostic results from clinicians.The experimental results show that the SVM classification model trained by SVM-InfFS feature selection method can identify patients more accurately and effectively,and the accuracy is as high as 82.5%.3.Clinical application of radiomics in differentiating renal cystic lesions.Firstly,the ReliefF-based forward selection algorithm(SVM-ReliefF)was applied to select proper features,then extreme learning machine was trained to construct classified model.Meanwhile,the sensitivity,specificity,accuracy,false positive,false negative and AUC used as evaluation indicators to evaluate the classification model,and the classification accuracy of different classifiers is compared and analyzed.The experimental results show that the ELM classification model trained by SVM-ReliefF feature selection method can identify patients more accurately and effectively,and the accuracy is as high as 83.1%.