Related Experiment Video
Updated: Aug 5, 2026
Metagenomic Next-Generation Sequencing of Cerebrospinal Fluid for the Detection of Central Nervous System Pathogens
Published on: April 17, 2026
A predictive model for central nervous system infections in children based on machine learning and clinical
Bin Zhou1, Feng Liang2, Yukun Huang3
1Department of Infectious Diseases, Children's Hospital of Fudan University (Xiamen Branch), Xiamen Children's Hospital, Xiamen, Fujian, China.
Insights
This study developed a machine learning model to differentiate pediatric Central Nervous System Infection (CNSI) from Febrile Seizures (FS) using initial clinical data, aiding early diagnosis.
Area of Science:
- Pediatric Neurology
- Medical Informatics
- Machine Learning in Medicine
Background:
- Pediatric Central Nervous System Infection (CNSI) symptoms overlap with Febrile Seizures (FS), complicating early diagnosis.
- Existing machine learning models often focus on general infection risk, not specific CNSI vs. FS differentiation using initial data.
Purpose of the Study:
- To create an exploratory machine learning model for early risk stratification to distinguish pediatric CNSI from FS.
- To identify key predictive clinical factors for differentiating CNSI from FS in children.
Main Methods:
- Retrospective analysis of 140 children hospitalized with fever and convulsions.
- Feature selection using Random Forest and XGBoost after univariate screening and multicollinearity handling.
- Model performance evaluation using Area Under the ROC Curve (AUC) and interpretability assessment with SHAP.
Main Results:
- Ten core features were identified, including Calcium, Lymphocyte Percentage, Albumin, RBC count, Oxygen Saturation, Lactic Acid, Neutrophil Percentage, Babinski sign, Headache, and EEG findings.
- Support Vector Machine (SVM) achieved an AUC of 0.864.
- Calcium, Oxygen Saturation, and Lymphocyte Percentage were identified as significant predictors by SHAP analysis.
Conclusions:
- An exploratory model was developed to differentiate pediatric CNSI from FS using readily available initial diagnostic variables.
- The model offers preliminary insights for early risk stratification of pediatric CNSI.
- Further validation via multicenter prospective studies is needed to confirm generalizability and clinical utility.
Background:
Early manifestations of pediatric Central Nervous System Infection (CNSI) lack specificity and are difficult to distinguish from Febrile Seizures (FS). Previous machine learning research has primarily focused on general infection risk stratification, with few studies on discriminative models for pediatric CNSI using initial clinical data.
Objective:
To construct an exploratory machine learning-based model for early risk stratification to differentiate pediatric CNSI from FS using single-center retrospective data and to analyze key predictive factors.
Methods:
Children hospitalized with fever and convulsions between January 2023 and December 2025 were enrolled. Initial clinical features and laboratory test results were used for univariate screening. After multicollinearity handling (Spearman correlation, |r| ≥ 0.70) and feature importance ranking via Random Forest (RF) and Extreme Gradient Boosting (XGBoost) algorithms, a union set of 10 variables was selected for modeling. The performance of multiple algorithms was compared based on the Area Under the Receiver Operating Characteristic Curve (AUC). Model interpretability was assessed using a nomogram and Shapley Additive exPlanations (SHAP).
Results:
A total of 140 children were included (28 in the CNSI group, 112 in the FS group). 10 core features were selected, including: Calcium (Ca) concentration, Lymphocyte Percentage (L%), Serum Albumin (ALB) level, Red Blood Cell (RBC) count, Oxygen Saturation (SO2), Lactic Acid (LAC), Neutrophil Percentage (N%), Babkinki sign, Headache, and Electroencephalogram (EEG) findings. Among the algorithms, Support Vector Machine (SVM) achieved a relatively high AUC of 0.864 (95% CI: 0.625-1.000) in the internal test set. SHAP analysis indicated that Ca, SO2, and L% contributed significantly to the model.
Conclusion:
This study developed an exploratory discriminative model for differentiating pediatric CNSI from FS using variables available at initial diagnosis. The model provides preliminary insights for early risk stratification of pediatric CNSI, though its generalizability and clinical utility require further validation through multicenter prospective external studies.