Automatic Prediction of Paediatric Cardiac Output From Echocardiograms Using Deep Learning Models

Area of Science:

• Cardiology
• Medical Imaging
• Artificial Intelligence

Background:

• Cardiac output (CO) perturbations are common, leading to significant morbidity and mortality.
• Accurate CO assessment is vital for treatment guidance in anesthesia and critical care.
• Current CO measurement methods are challenging, even for experts, highlighting the need for advanced solutions.

Purpose of the Study:

• To develop and evaluate a novel deep learning model for accurate and rapid pediatric cardiac output estimation.
• To modify an existing adult deep learning model (EchoNet-Dynamic) for pediatric left ventricular outflow tract diameter prediction.
• To create a new deep learning approach for velocity time integral estimation to enable automatic CO prediction.

Main Methods:

• Paediatric echocardiograms from normal and dilated cardiomyopathy groups were reviewed.
• A modified EchoNet-Dynamic model was retrained on pediatric data for LVOT diameter prediction.
• A novel deep learning model was developed for VTI estimation, combined with LVOT prediction for automated CO calculation.
• Model performance was evaluated against expert measurements using RMSE, MAE, MAPE, and R².

Main Results:

• The model estimated cardiac index (CI) with an R² of 0.755, RMSE of 0.389 L/min/m², MAE of 0.321 L/min/m², and MAPE of 10.8%.
• Bland-Altman analysis revealed a bias of +0.14 L/min/m² and 95% limits of agreement from -0.58 to 0.86 L/min/m² for CI estimation.
• The developed model demonstrated strong correlation with ground truth CO measurements, showing a bias of 0.17 L/min.

Conclusions:

• The AI model accurately estimates pediatric cardiac output, outperforming many existing methods.
• Model pretraining facilitated accurate estimation even with a limited dataset.
• Potential applications include real-time bedside CO calculation, identification of low-CO states, and monitoring treatment responses.