Early risk stratification of adverse progression in pediatric Mycoplasma pneumoniae pneumonia using a multicenter CT radiomics model.

OBJECTIVE

This study sought to establish and externally validate an interpretable CT-based radiomics framework for early prediction of adverse disease progression in children with Mycoplasma pneumoniae pneumonia (MPP).

METHODS

In this multicenter retrospective study, 419 children with MPP from three hospitals were assigned to training (n = 296), testing (n = 59), and validation (n = 64) cohorts. Adverse progression was defined as refractory MPP, plastic bronchitis, or both. Radiomics features were extracted from pulmonary lesions and selected using SelectKBest and least absolute shrinkage and selection operator (LASSO) regression. Three random forest models, namely the clinical-imaging, radiomics, and integrated models, were evaluated by receiver operating characteristic (ROC) analysis, calibration, decision curve analysis, and pairwise comparisons.

RESULTS

Among the 419 included children, 70 developed adverse progression. In the training cohort, D-dimer, white blood cell count, systemic immune-inflammation index, lobar atelectasis, and number of involved lung lobes were significantly associated with progression (all p < 0.05). In the external validation cohort, the integrated model (AUC, 0.828; 95% CI, 0.645-0.957) achieved an accuracy of 0.906, F1 score of 0.727, sensitivity of 0.727, and specificity of 0.943, outperforming the Clinical-Imaging model (AUC, 0.736; 95% CI, 0.514-0.913; p = 0.013) and radiomics model (AUC, 0.813; 95% CI, 0.628-0.950; p = 0.043).

CONCLUSION

An integrated model incorporating inflammatory indices, CT findings, and radiomics features showed promise for early risk stratification for adverse progression in pediatric MPP.