Machine learning models based on chest computed tomography for identifying plastic bronchitis in children with
Yanjie Xu1,2, Zhiling Liu3, Jingshuo Li4
1Department of Radiology, Shandong Provincial Qianfoshan Hospital, Shandong University, Jinan, China.
Background:
Early identification of plastic bronchitis (PB) in Mycoplasma pneumoniae pneumonia (MPP) is very important, as it may help to initiate appropriate treatment early. We aimed to establish a machine learning (ML) model integrating clinical risk factors and chest computed tomography (CT) features to predict PB in children with MPP complicated with lung consolidation.
Methods:
This retrospective study collected 777 children diagnosed with MPP complicated with lung consolidation from three clinical centers between January 2019 and October 2024. Among them, 280 developed PB. The patients were divided into training set [n=373, The Second Qilu Hospital of Shandong University (Center 1)], test set [n=221, Shandong Provincial Hospital Affiliated to Shandong First Medical University (Center 2)], and validation set [n=183, The First Affiliated Hospital of Shandong First Medical University (Center 3)]. The whole lung area on chest CT was defined as region of interest. The least absolute shrinkage and selection operator regression was used to select the most significant radiomics features. Univariate and multivariate logistic regression analyses were conducted to develop a multifactorial model combining radiological and clinical risk factors. Model performance was evaluated using the receiver operating characteristic curve, and clinical usefulness of the models was assessed through decision curve analysis (DCA).
Results:
Seven radiomics features and pleural effusion were used to develop prediction models. The multifactorial model demonstrated the highest area under the curve values of 0.809 [95% confidence interval (CI): 0.763-0.855], 0.770 (95% CI: 0.702-0.839), and 0.831 (95% CI: 0.764-0.897) in the training set, the test set, and the validation set, respectively, which were significantly higher than those of CT-only or clinical-only models. The calibration curves indicated that the multifactorial model achieved superior agreement between predicted and observed outcomes, and DCA showed that it provided a greater net clinical benefit.
Conclusions:
The multifactorial model enabled early, reliable identification of PB in children with MPP prior to fiberoptic bronchoscopy.
