Application study of CT radiomics features in short-term prognosis of patients with acute craniocerebral injury
Objective:To explore the predictive effect of CT radiomics features on short-term prognosis of patients with acute cranio-cerebral injury(ACI).Methods:Clinical data of 100 ACI patients treated in a hospital from March 2021 to June 2022 were retrospec-tively analyzed.The patients were divided into a good prognosis group(n =57)and a poor prognosis group(n =43)after 3 months'treatment based on the result of Glasgow coma scale(GCS).According to the CT radiomics parameters of the patients on admission,regions of interest(ROIs)of craniocerebral injury were drawn with ITk-snap and CT radiomics features were extracted.The features were screened with AUC analysis and LASSO regression of 5-fold cross validation method.Spearman correlation analysis was used to e-liminate redundant features.Predictive model based upon the SVM was set up accordingly.Another 30 ACI patients were prospectively chosen on admission.Their prognosis was appraised with SVM model and Helsinki CT score and predictive effect was compared using ROC.Results:SVM model analysis indicated that the first 6 features influencing the short-term poor prognosis were intensity of short in-terval high grayscale value,grayscale variance,grayscale nonuniformity,small area high gray level emphasis,nonuniformity in normal-ized size area and extensive protruding value.ROC analysis showed that SVM model was superior to Helsinki CT score in predicting effect.For the SVM model,its AUC value was 0.921,the optimum cutoff value was 0.442,the corresponding sensitivity and specifici-ty were respectively 0.884 and 0.860.Conclusion:Based upon CT radiomics features,the SVM model for the short-term prognosis of ACI patients possesses good predictive effect,which can provide clinical basis for the poor prognosis of death and recurrence in ACI pa-tients after treatment.
Acute craniocerebral injuryRadiomicsPrognosisPredictive model
万方数据
维普
