Artificial Intelligence Automation of Echocardiographic Measurements

Area of Science:

• Cardiology
• Medical Imaging
• Artificial Intelligence

Background:

• Accurate echocardiographic measurements are vital for cardiovascular disease diagnosis and monitoring.
• Manual echocardiogram analysis is time-consuming and prone to imprecision.
• Artificial intelligence (AI) offers a solution to automate these measurements, reducing clinician workload.

Purpose of the Study:

• To develop and validate open-source deep learning semantic segmentation models.
• To automate the measurement of 18 anatomic and Doppler parameters in echocardiography.
• To assess the accuracy and precision of AI-driven measurements against expert sonographer assessments.

Main Methods:

• Trained deep learning models on a large dataset (877,983 measurements from 155,215 studies) from Cedars-Sinai Medical Center (CSMC).
• Validated model performance using held-out data from CSMC and an external dataset from Stanford Healthcare (SHC).
• Compared AI-generated measurements with sonographer measurements for accuracy and precision across B-mode and Doppler parameters.

Main Results:

• Developed EchoNet-Measurements, an open-source AI model for automated echocardiographic annotation.
• Achieved high accuracy in measurements, with strong performance on both internal (CSMC) and external (SHC) validation datasets.
• Demonstrated consistent performance across diverse patient demographics and equipment vendors, highlighting robustness.

Conclusions:

• EchoNet-Measurements provides highly accurate automated echocardiographic quantification.
• The AI model has significant potential to assist clinicians in echocardiography workflows.
• This open-source tool serves as a foundation for future advancements in AI for echocardiography.