Detect Neovascularization Of Carotid Plaque On Unenhanced Mri By Texture Features

Objective1.By combining texture analysis with unenhanced MRI,attempting to identify intraplaque neovascularization without contrast agent,in order to establish a diagnostic model of intraplaque neovascularization.In addition,combining with traditional clinical serum indicators to build a clinical-radiomics diagnostic model to provide assistance for exploring the feasibility of diagnosis intraplaque neovascularization on unenhanced MRI.2.Furthermore,based on texture analysis,the possibility of identifying different grades of intraplaque neovascularization on unenhanced MRI was explored,which opened a new way for quantitative identification of high-grade and low-grade intraplaque neovascularization.Methods1.Patients with carotid plaque were collected,and all the subjects underwent contrast enhanced ultrasound and carotid MRI(T1WI,T2 WI and contrast enhancement-T1 weighted images(CE-T1WI)).According to the results of contrast enhanced ultrasound,94 carotid plaques were divided into intraplaque neovascularization group(IPN group)and non-intraplaque neovascularization group(non-IPN group).The largest plaque was selected on the MRI,and the region of interest(ROI)of plaque was manually sketched by an experienced radiologist,and the features from the ROI were extracted.All features were reduced and selected by intragroup correlation coefficient(ICC),random forest and correlation detection.Finally,the statistical differences of important features between IPN group and non-IPN group were compared,and a diagnostic model was constructed.Combined with the traditional clinical serum indicators respectively,the comprehensive clinical-radiomics diagnosis model was constructed.2.According to the contrast enhanced ultrasound,65 plaques with intraplaque neovascularization were divided into low-grade group(1-2 grade)and high-grade group(3-4 grade).The three-dimensional texture features were extracted from the ROI and selected based on the ICC,recursive feature elimination(RFE)and correlation detection to construct a diagnostic model of different grades of intraplaque neovascularization.Results1.According to the results of contrast enhanced ultrasound,94 subjects with carotid plaques were divided into IPN group(n= 65)and non-IPN group(n= 29).The AUC of CE-T1 WI for identifying intraplaque neovascularization was 0.733,the sensitivity was 84.620%,and the specificity was 62.070%.However,intraplaque neovascularization could not be recognized on unenhanced MRI(T1WI and T2WI).2.In this experiment,a total of 558 texture features were extracted.Firstly,165 low reproducible texture features with ICC<0.75 were excluded.Then the directionrelated features of the remaining 393 features were integrated and averaged,and 70 texture features were left.Then,16 texture features were left after random forest dimensionality reduction.Finally,the high collinearity features were eliminated,and five important and independent texture features were obtained.These five features include three features from Co-occurrence matrix(T1_DifVarnc,T2_SumAverg and T2_Entropy),Absolute gradient(T2_GrSkewness)and one feature from Autoregressive model(T2_Teta 2).3.Except for the T2_Entropy,significant differences were in other four features between the IPN group and the non-IPN group(P<0.05).When these four features were incorporated into the logistic regression equation,and it was found that T1_DifVarnc,T2_SumAverg and T2_Teta 2 satisfied the equation.Then,these three texture features were combined to construct a diagnostic model of intraplaque neovascularization.In addition,the AUC of the model was 0.887,the sensitivity was 89.230%,and the specificity was 75.860%.4.A clinical-radiomics diagnostic model was constructed by combining the diagnostic model of T1 WI + T2 WI with clinical serum indicators.The results showed that the diagnostic performance of the comprehensive model was significantly improved and texture analysis + T1 WI + T2 WI + LDL achieved the strongest diagnostic ability with an AUC of 0.934.5.In the high-grade and low-grade intraplaque neovascularization group,through feature extraction and multi-step dimensionality reduction,it was found that statistical different features T1_GLCM_Contrast and T2_GLCM_Energy.The diagnostic model constructed by these two features was AUC= 0.721,and its corresponding sensitivity and specificity were 84.210% and 59.260%,respectively.Conclusion1.Texture analysis can identify intraplaque neovascularization on unenhanced MRI,in which T1_DifVarnc,T2_SumAverg and T2_Teta2 can be used as good indicators to reflect the presence or absence of intraplaque neovascularization and play an important role in the diagnosis of intraplaque neovascularization.2.Combined with serum indicators to construct a clinical-radiomics model,the construction of clinical-radiomics comprehensive diagnostic model can better improve the diagnostic ability and provide the possibility for further clinical detection of intraplaque neovascularization.3.The application of texture analysis to identify high-grade and low-grade intraplaque neovascularization provided more experimental evidence.