Predictive activity of pulmonary cystic echinococcosis based on CT radiomic classifier model
10.3760/cma.j.cn112149-20240117-00030
- VernacularTitle:基于CT放射组学分类器模型预测肺囊型包虫病活性的研究
- Author:
Yaohui YU
1
;
Yuan ZHAO
;
Yan LI
;
Xuehong LU
;
Yang JING
;
Yan XING
Author Information
1. 新疆医科大学第一附属医院影像中心，乌鲁木齐 830054
- Keywords:
Echinococcosis, pulmonary;
Tomography, X-ray computed;
Machine learning;
Predictive model;
Radiomics
- From:
Chinese Journal of Radiology
2024;58(10):1050-1055
- CountryChina
- Language:Chinese
-
Abstract:
Objective:To explore the value of the classifier model based on CT radiomic characteristics in predicting the activity of pulmonary cystic echinococcosis (CE).Methods:The study was designed as cross-sectional. A retrospective analysis was performed on 81 patients diagnosed with pulmonary CE in the First Affiliated Hospital of Xinjiang Medical University from January 2010 to October 2020. The Python function divided 17 cases into an internal verification group and 64 cases into a training group with a ratio of 2∶8. In addition, 16 patients diagnosed with pulmonary CE from the Fourth Affiliated Hospital of Xinjiang Medical University from October 2020 to 2024 were included in the external validation group. All patients underwent CT examination, and radiomics features were extracted using Radcloud platform of Huimedi Huiying. The intraclass correlation coefficient was performed on the features, then feature screening was performed using the SelectKBest method, variance thresholding method, and least absolute shrinkage and selection operator. Finally, three classifiers (including support vector machine (SVM), K-neighborhood (KNN), and logistic regression (LR)) were used to build the models. The receiver operating characteristic curve and area under the curve (AUC) were used to evaluate the model′s efficiency.Results:Among 81 patients with lung CE, 58 were male, and 23 were female. twenty-eight lesions were active and 53 were inactive. A total of 11 optimal features were selected. Based on the selected features, the SVM classifier model, KNN classifier model, and LR classifier model were established. The KNN classifier model in the training group had the highest AUC value (0.93) and the highest specificity (0.98) in predicting lung CE activity. In the internal validation group, the SVM classifier model had the highest AUC value (0.92) and the highest specificity (0.91) in predicting lung CE activity. The LR classifier model performed best with the highest AUC of 0.85 for predicting lung CE activity in the external validation group, and the specificity of the three models was 0.92.Conclusion:The classifier model established based on CT radiomic features has a certain value in predicting lung CE activity, and may be helpful in clinical decision-making.
