EchoAI-Peds achieved a macro-averaged AUROC of 0.91 on the internal test set and 0.89 on the external test set, significantly outperforming adult-based echocardiography foundation models.
Observational (n=10,168)
Yes
Does EchoAI-Peds improve the detection of congenital heart defects and abnormalities in pediatric echocardiograms compared to adult-based models?
A multi-task deep learning model specifically tailored for pediatric echocardiography demonstrated high diagnostic accuracy for detecting congenital heart defects and outperformed adult-based foundation models.
Effect estimate: AUROC 0.91 (95% CI 0.90-0.92)
Absolute Event Rate: 0.91% vs 0.58%
p-value: p=<0.001
Background: Congenital heart defects afflict nearly 1% of all births worldwide. While deep learning algorithms have shown significant promise in automating and improving adult echocardiography analysis, similar progress has not been observed in pediatric echocardiography. Specifically, existing pediatric-based models are limited to single tasks and specific echocardiographic views. To address this, we introduce EchoAI-Peds, the first multi-task deep learning model for pediatric echocardiography. Our model was developed using the most comprehensive set of pediatric labels to date and is designed to integrate information from multiple echocardiographic views simultaneously. Methods: A video-based vision transformer was trained to simultaneously detect 28 congenital heart defects, structural and functional abnormalities, repairs, and interventions directly from complete pediatric echocardiography studies with multiple videos. During inference, our model integrates information from all available views to produce unified study-level predictions. Our model was developed using over 700,000 videos derived from more than 11,000 studies at Stanford Medicine. Model efficacy was tested on an internal held-out dataset. In addition, model generalizability was tested on a spatially and temporally distinct patient cohort at the Children's Hospital of Philadelphia. Results: Our model achieved macro-averaged AUROC values of 0.91 (95% CI: 0.90-0.92) and 0.89 (95% CI: 0.88-0.90) on the internal and external test sets, respectively. Moreover, our model significantly outperformed adult-based echocardiography foundation models trained on substantially larger datasets (p < 0.001). Finally, our model demonstrated robust performance across patient age, patient sex, and studies with varying number of videos. Conclusions: Our findings demonstrate the remarkable potential for multi-task deep learning models to aid the interpretation of pediatric echocardiograms. In addition, our results underscore the need for models that are specifically tailored to pediatric populations.
Mrudang et al. (Thu,) conducted a observational in Congenital heart defects and pediatric structural/functional cardiac abnormalities (n=10,168). EchoAI-Peds (Multi-Task Deep Learning Model) vs. Adult-based echocardiography foundation models (EchoCLIP and EchoPrime) was evaluated on Macro-averaged AUROC for detecting 28 clinical findings on the internal test set (AUROC 0.91, 95% CI 0.90-0.92, p=<0.001). EchoAI-Peds achieved a macro-averaged AUROC of 0.91 on the internal test set and 0.89 on the external test set, significantly outperforming adult-based echocardiography foundation models.