Automated 2-D and 3-D Left Atrial Volume Measurements Using Deep Learning

Area of Science:

• Cardiology
• Medical Imaging
• Artificial Intelligence

Background:

• Echocardiography is vital for assessing left atrial (LA) volume.
• Current LA volume measurement methods are often manual or semi-automated, leading to variability.
• There is a need for accurate, efficient, and reproducible LA volume assessment.

Purpose of the Study:

• To develop and validate a fully automated, real-time method for measuring LA volume from 2-D and 3-D echocardiography.
• To achieve accuracy comparable to expert assessments while reducing operator variability and time.
• To provide valuable temporal LA volume curve information in the echocardiography laboratory.

Main Methods:

• Developed an automated pipeline using a network for end-systole (ES) time point identification.
• Employed robust 2-D and 3-D U-Nets for image segmentation.
• Utilized datasets of 789 2-D images and 286 3-D recordings, exploring recurrent networks and pseudo-labeling for volume curve estimation.

Main Results:

• Baseline automated measurements showed low average volume differences: 2.9 mL for 2-D and 7.8 mL for 3-D compared to manual methods.
• Pseudo-labeling enhanced the robustness of volume curves and improved ES measurement, especially with limited data.
• The system processed real-time data from a clinical ultrasound scanner.

Conclusions:

• Automated LA volume estimation shows significant potential for clinical application.
• The developed prototype provides real-time, accurate LA volume assessment.
• This technology can improve efficiency and consistency in echocardiography assessments.