Real-Time Automatic M-Mode Echocardiography Measurement With Panel Attention

Area of Science:

• Cardiology
• Medical Imaging
• Artificial Intelligence

Background:

• M-mode echocardiography is crucial for cardiac measurements but is time-consuming and prone to accuracy variations.
• Current diagnostic methods lack efficiency and consistency, necessitating automated solutions.
• Deep learning offers potential for developing accurate and efficient automated diagnostic schemes.

Purpose of the Study:

• To develop an automated scheme for real-time M-mode echocardiography (RAMEM) to enhance diagnostic accuracy and efficiency.
• To introduce the first M-mode echocardiogram dataset (MEIS) for consistent training and evaluation.
• To propose an efficient algorithm (AMEM) for automated M-mode echocardiography measurements.

Main Methods:

• Developed RAMEM, an automated real-time M-mode echocardiography scheme.
• Created MEIS, a novel dataset of M-mode echocardiograms.
• Proposed panel attention embedding with UPANets V2 for real-time instance segmentation (RIS) to improve object detection.
• Introduced AMEM, an efficient algorithm for automated M-mode echocardiography measurement.

Main Results:

• RAMEM demonstrated superior performance compared to existing RIS schemes and human performance on the MEIS dataset.
• The proposed panel attention embedding enhanced big object detection in echocardiograms.
• The system achieved competitive results on the PASCAL 2012 SBD benchmark.

Conclusions:

• RAMEM offers a promising automated solution for M-mode echocardiography, addressing current diagnostic limitations.
• The developed MEIS dataset and AMEM algorithm contribute to advancing automated cardiac diagnostics.
• This work highlights the potential of deep learning in improving the speed and accuracy of echocardiogram analysis.