| Objective:To explore the diagnostic value of enhanced CT image texture analysis in the differentiation of benign and malignant small renal masses.Materials and methods:Fifty-one patients with benign small renal mass(38 with fat-poor angiomyolipoma and 13 with oncocytoma)and 65 patients with malignant small renal mass(all with clear cell renal cell carcinoma)were retrospectively collected in our hospital.All cases were confirmed by surgery and pathology.The CT imaging features of the lesions in the two groups were analyzed and evaluated.The maximum axial slice images of the lesions in the cortical and parenchymal stages were selected,and MaZda software was used to extract texture parameters,including the grey-level histogram,grey-level cooccurrence matrix,grey-level run-length matrix,grey-level absolute gradient,autoregressive model and wavelet transform.Feature selection algorithms were included(Fisher coefficient[Fisher],mutual information[MI],probability of classification error and average correlation coefficient[POE+ACC]).The three texture feature selection algorithms were used to screen the best texture parameters.The independent samples t test,Mann-Whitney U test or χ2 test between the two groups were used to screen out the imaging features and texture parameters with statistically significant differences.Consistency analysis of texture parameters was carried out using Intra-class correlation coefficients.Spearman correlation analysis was used to assess differences in texture parameters between the two groups.Based on the CT image features with statistically significant differences,the independent influencing factors between benign and malignant groups were analyzed by multivariate logistic regression.The models were built using logistic regression as the classifier and validated by five-fold cross-validation.The ROC curve was drawn,the classification accuracy and AUC value were calculated,and the diagnostic performance of each model was evaluated and compared.Results:There were statistically significant differences in the density of lesions,pseudocapsule,cystic necrosis,enhancement uniformity,enhancement degree,fast in and fast out,delayed enhancement and 10 groups of texture parameters.The results of logistic regression analysis based on CT image features showed that enhancement uniformity and delayed enhancement were independent influencing factors in distinguishing benign and malignant small renal masses.The average classification accuracy of the CT image feature prediction model was 0.67±0.14,and the average AUC was 0.65±0.11.The average classification accuracy of the CT texture parameter prediction model was higher,0.74±0.07;the AUCs of the five-fold cross-validation were 0.72,0.66,0.78,0.73,and 0.83,the average AUC was 0.74±0.06,and the sensitivity was 0.75±0.10,The specificity was 0.73±0.08.Conclusions:Enhanced CT image texture analysis had a good diagnostic value for the identification of benign and malignant small renal masses,and might be helpful for clinical decision-making and prognosis evaluation.Objective:To explore the application value of texture parameters and imaging features of multiphase CT images in risk stratification of small renal cell carcinoma.Materials and methods:A total of 125 cases of renal clear cell carcinoma(≤4cm)confirmed by postoperative pathology in our hospital were retrospectively collected,and they were divided into high-risk group(60 cases)and medium-low risk group(65 cases)according to postoperative pathological characteristics.The CT imaging features of the lesions in the two groups were analyzed,and MaZda software was used to analyze the texture of the plain scan,cortical and parenchymal thin-layer images.The optimal texture parameters were screened by Fisher,POE+ACC and MI.The independent samples t test,Mann-Whitney U test or χ2 test between the two groups were used to screen out the imaging features and texture parameters with statistically significant differences.Consistency analysis of texture parameters was carried out using Intra-class correlation coefficients.Spearman correlation analysis was used to assess differences in texture parameters between the two groups.Based on CT texture parameters and CT imaging features,and combined with random forest machine learning algorithm,a total of six prediction models were established,including plain scan-CT texture parameter model,cortical phase-CT texture parameter model,parenchymal phase-CT texture parameter model,multi-phase-CT texture parameter model,CT imaging feature model,and imaging feature combined multi-phase texture parameter model.The average classification accuracy of each model in identifying high-risk small renal carcinoma was calculated,and the ROC curve analysis was performed to obtain the AUC value.Results:Four CT imaging features were statistically different between the two groups,including the longest diameter of the tumor,CT value of plain scan,lesion boundary and morphology.The differences between the 12 groups of CT texture parameters were statistically significant,including 5 groups from plain scan images,4 groups from cortical phase images,and 3 groups from parenchymal phase images.Among the single-phase texture parameter prediction models,the average classification accuracy of the plain scan-CT texture parameter model was higher,at 0.66±0.10.The average classification accuracy of the multi-phase-CT texture parameter prediction model was 0.67±0.07,and the average AUC was 0.68±0.07.The mean classification accuracy of the CT imaging feature prediction model was 0.66±0.16,and the mean AUC was 0.67±0.17.Among the six models,the imaging feature combined multi-phase texture parameter prediction model had the highest average classification accuracy,which was 0.77±0.10;the average AUC of this model was 0.78±0.10,the sensitivity was 0.78±0.06,and the specificity was 0.7±0.19.Conclusions:1.The diagnostic performance of conventional CT imaging features and image texture parameters of a single CT scan phase in the identification of high-risk small renal carcinoma was general,and the combination of multi-phase image texture parameters couldn’t significantly improve the diagnostic performance.2.Multiphase CT image texture parameters combined with image features could better stratify the risk of small renal carcinoma,which was expected to provide a new method for the selection of individualized treatment strategies for patients. |
PDF Full Text Request