Leveraging Comprehensive Echo Data to Power Artificial Intelligence Models for Handheld Cardiac Ultrasound

Area of Science:

• Artificial Intelligence in Medicine
• Cardiovascular Imaging
• Deep Learning Applications

Background:

• Echocardiography is crucial for assessing cardiac function.
• Handheld cardiac ultrasound (HCU) offers portability but requires robust analytical tools.
• Deep learning presents an opportunity to enhance echocardiographic analysis.

Purpose of the Study:

• To develop an end-to-end deep learning framework for analyzing echocardiographic videos.
• To estimate left ventricular ejection fraction (LVEF), patient age, and classify patient sex.
• To evaluate the framework's performance on both traditional transthoracic echocardiography (TTE) and handheld cardiac ultrasound (HCU) data.

Main Methods:

• Trained deep learning models on a large retrospective TTE dataset (6432 studies).
• Validated models on internal (1369 studies) and external retrospective TTE datasets (4359 studies across three sites).
• Applied models to a prospective dataset including HCU and TTE videos (625 patients).

Main Results:

• Strong performance on retrospective TTE data for LVEF estimation (RMSE 6.53-6.95%) and classification (AUC 0.962-0.980).
• Accurate age estimation (RMSE 9.44%) and sex classification (AUC 0.882) from TTE.
• Comparable performance for HCU and TTE in prospective analysis for LVEF, age, and sex.

Conclusions:

• Retrospective TTE data effectively trains deep learning models for HCU applications.
• Demonstrates the feasibility of obtaining focused diagnostic images with handheld devices.
• Highlights the potential of deep learning to standardize and improve cardiac ultrasound analysis.