A Multitask Deep Learning Model for Pediatric Echocardiography Analysis

Area of Science:

• Artificial Intelligence
• Medical Imaging
• Pediatric Cardiology

Background:

• Congenital heart defects affect approximately 1% of newborns globally.
• Existing deep learning models for echocardiography are often limited to single tasks and specific views, particularly in pediatrics.
• There is a need for advanced AI tools tailored to the complexities of pediatric echocardiography.

Purpose of the Study:

• To introduce EchoAI-Peds, a novel multitask deep learning model for pediatric echocardiography.
• To develop a model capable of integrating information from multiple echocardiographic views simultaneously.
• To improve the accuracy and efficiency of diagnosing congenital heart defects in children.

Main Methods:

• A video-based vision transformer was trained to detect 28 different pediatric cardiac abnormalities from complete echocardiography studies.
• The model utilized a comprehensive dataset of over 700,000 videos from more than 12,000 pediatric studies.
• Training and validation were performed on extensive datasets from Stanford Medicine, with generalizability tested on data from the Children's Hospital of Philadelphia.

Main Results:

• EchoAI-Peds achieved high performance with macroaveraged area under the receiver operating characteristic curve (AUROC) values of 0.91 (internal) and 0.89 (external).
• The model significantly outperformed established adult-based echocardiography foundation models on both internal and external test sets.
• Performance remained robust across diverse patient demographics (age, sex) and varying study complexities (number of videos).

Conclusions:

• Multitask deep learning models hold significant potential for assisting in the interpretation of pediatric echocardiograms.
• The development of specialized AI models tailored to pediatric populations is crucial for advancing cardiac diagnostics.
• EchoAI-Peds represents a significant step towards more accurate and comprehensive AI-driven analysis of pediatric cardiac imaging.