Objective To investigate the value of a prediction model combining blood indices and CT radiomic features in diagnosing refractory Mycoplasma pneumoniae pneumonia (RMPP) progressing to pulmonary consolidation.

Methods This study was a single-center retrospective cohort study that collected clinical and imaging data from 120 hospitalized pediatric patients aged 1–18 years diagnosed with Mycoplasma pneumoniae pneumonia at The First Affiliated Hospital of Bengbu Medical University, including 60 patients with RMPP and 60 with general Mycoplasma pneumoniae pneumonia (GMPP), all with pulmonary consolidation confirmed by CT imaging. The patients were randomly divided into a training set and a test set in a 7∶3 ratio. A three-dimensional region of interest (ROI) was delineated using the Maximum cross-sectional area of the entire lesion on CT images. Optimal radiomic features were obtained through extraction and screening, and optimal clinical features were selected concurrently. Three prediction models for RMPP progressing to pulmonary consolidation were established: a clinical prediction model, a radiomic prediction model, and a comprehensive prediction model combining clinical and radiomic features. The ROC curve was used to evaluate the predictive performance of the three models.

Results The platelet (PLT) count, C-reactive protein (CRP) level, lactate dehydrogenase (LDH) level, and D-dimer level were significantly higher in the refractory group than in the general group 348.5 (279.5, 431) vs 270 (225.5, 369.5)、28.44 (11.98, 63.57) vs 10.63 (5, 27.67)、10.63 (5, 27.67) vs 291.5 (255.3, 334.5)、0.89 (0.54, 2.63) vs 0.28 (0.21, 0.43), with statistically significant differences (P ＜ 0.05). The combined prediction model exhibited the best predictive performance, with an area under the curve (AUC) of 0.968 (0.934–1.000) for the training set and 0.858 (0.736–0.980) for the validation set, significantly higher than the radiomic model 0.931 (0.885–0.987)、0.713 (0.539–0.886) and the clinical model 0.951 (0.911–0.990)、0.923 (0.840–1.000).

Conclusion The combined prediction model integrating clinical features (CRP, D-dimer, LDH, PLT) and radiomic features has higher clinical significance in predicting RMPP progressing to pulmonary consolidation and can provide guidance for developing individualized treatment plans in clinical practice.