Predictive value of CT texture analysis for early enlargement of hypertensive intracerebral hemorrhage / 实用放射学杂志

ABSTRACT

Objective To explore the predictive value of CT image texture analysis for early enlargement of hypertensive intracerebral hemorrhage.Methods One hundred and eight patients with hypertensive intracerebral hemorrhage were divided into enlarged hematoma group (positive group)and non-enlarged hematoma group (negative group),according to whether the volume of hematoma on 24 h follow up CT scan was more than 30% or 6 mL of the baseline CT.Phillis Radiomics Tool V93 software was used to segment the hematoma on CT plain scan images of two groups,four features of first-order and three of gray-level co-occurrence matrix (GLCM),thirteen of gray-level size zone matricx (GLSZM)and eleven of gray-level run-length matricx (GLRLM)were obtained.The differences of thirty-one texture features between the two groups were compared.The ROC curves of the features with statistical differences were analyzed.The independent predictors of early enlargement of intracerebral hemorrhage were screened by Logistic multivariate regression model.Results Among the one hundred and eight patients,twenty-eight were positive group and eighty were negative group.Skewness and long run low gray-level emphasis (LRLGE)in positive group were significantly higher than those in negative group (P＜0.05).There was no significant difference in the remaining twenty-nine features between the two groups (P＞0.05).ROC curve analysis showed that the AUC of Skewness,LRLGE and their combined diagnosis were 0.634,0.814 and 0.828,respectively.The independent variables were screened by stepwise regression analysis.The LRLGE (OR＝1.238,95%CI＝1.009－1.51 9,P＜0.05)was selected as the regression model, suggesting that LRLGE was an independent predictor of the early enlargement of intracerebral hemorrhage.Conclusion Texture analysis of CT images is helpful to predict the early enlargement of hypertensive intracerebral hemorrhage,and LRLGE based on GLRLM algorithm can be used as an independent predictor.