Using machine learning models based on cardiac magnetic resonance parameters to predict the prognostic in children...

Area of Science:

• Cardiology
• Medical Imaging
• Machine Learning

Background:

• Pediatric myocarditis can lead to adverse cardiac events.
• Accurate prognostic prediction is crucial for timely intervention.
• Cardiac magnetic resonance (CMR) offers detailed cardiac insights.

Purpose of the Study:

• To develop machine learning (ML) models for predicting pediatric myocarditis prognosis.
• To identify key cardiac magnetic resonance (CMR) parameters for prognosis prediction.
• To evaluate the performance of different ML models in this prediction task.

Main Methods:

• Retrospective analysis of 77 pediatric myocarditis patients.
• Utilized ultrasound, ECG, biomarkers, and CMR scans for data collection.
• Developed and compared four ML models: LR, RF, SVC, and XGBoost, evaluating performance using AUC.

Main Results:

• Late gadolinium enhancement (LGE), left ventricular ejection fraction (LVEF), and strain parameters (SAXGCS, LAXGLS) were key predictors.
• The logistic regression (LR) model achieved the highest prediction performance (AUC: 0.893).
• Reduced LVEF, SAXGCS, and LAXGLS, along with LGE, indicate a poor prognosis.

Conclusions:

• ML models, especially LR, show significant potential in predicting pediatric myocarditis prognosis.
• Identified CMR parameters provide valuable prognostic information.
• These findings can aid clinicians in making informed decisions and improving patient outcomes.