Objective To explore the predictive value of CT target scanning based radiomics and clinical features for pulmonary nodule malignancy.Methods The clinical and CT image data of 93 patients(102 pulmonary nodules)were retrospectively analyzed.The clinical and CT image features of 59 nodules in malignant group and 43 nodules in benign group were compared.The risk factors for pulmonary nodule malignancy were analyzed by multivariate logistic regression,and the clinical model was established.The features of target scan CT images were extracted and the radiomics model was established using logistic regression classifier.Seventy-one nodules in training group and 31 nodules in testing group were at a ratio of 7∶3.ROC curves were drawn to evaluate the predictive efficacy of the three models,which included clinical model,radiomics model and combined model,in training group and testing group for pulmonary nodule malignancy.Results CT image features including long diameter,short diameter,ratio of long-short diameter,density,shape,burr,air bronchial sign,pleural depression sign and vascular aggregation sign,were significantly different between malignant group and benign group(P＜0.05 or P＜0.01).Density of mixed ground glass or solid was an independent risk factor for pulmonary nodule malignancy(P＜0.01),which was a clinical model.The 10 image features with the highest weight coefficients were classified into radiomics model.The combined model based on CT target scanning showed a higher predictive efficacy,with AUC of 0.933(95％CI:0.880-0.987)in traning group and AUC of 0.885(95％CI:0.765-1.000)in testing group(P＜0.01).Conclusion CT target scanning based combined model of radiomics and clinical features has high predictive value for pulmonary nodule malignancy.
NETL
NSTL
万方数据
