Objective To discuss the value of imaging characteristics and clinical indicators of enhanced preoperative tomography(CECT)for predicting early postoperative recurrence of hepatocellular carcinoma(HCC).Methods Imaging and clinical data collected from 74 HCC subjects were randomly assigned to a training dataset(n=59)and a test dataset(n=15).ITK-SNAP was used to map the lesion manually.Pyradiomics was used for image feature extraction.The smallest absolute shrinkage and selection operator(LASSO)were used for dimension reduction.Finally,the recursive feature elimination(RFE)method was used to synthesize the characteristics of the omics image.The early recurrence prediction model of HCC was established by using Logistic regression(LR).The Logistic regression model was constructed by introducing clinical factors,and the nomogram was established by combining with the image omics score.Model performance was assessed using the 95%confidence interval(CI),AUC,SD,sensitivity,specificity and accuracy.Results The AUC of the clinical model on the training set was 0.745(95%CI:0.646～0.844).The AUC of the radiomic model was 0.805(95%CI:0.714～0.897),and the model performance was significantly improved to 0.846(95%CI:0.763～0.929)when the machine learning model was trained with image features and clinical indicators.Conclusion The model based on CT imaging combined with clinical characteristics can be used to predict early HCC recurrence.
维普
万方数据
