MCAL: An Anatomical Knowledge Learning Model for Myocardial Segmentation in 2-D Echocardiography

Area of Science:

• Medical Imaging
• Cardiovascular Ultrasound
• Artificial Intelligence in Medicine

Background:

• Accurate segmentation of the left ventricular (LV) myocardium in 2-D echocardiography is crucial for clinical assessments, including geometry measurement and index computation.
• Segmentation is challenging due to fuzzy boundaries caused by low image quality, and traditional methods using ground-truth labels offer limited feature enhancement.
• Existing approaches struggle with effective feature enhancement for 2-D echocardiography segmentation.

Purpose of the Study:

• To develop an advanced training strategy for improving the accuracy of left ventricular myocardium segmentation in 2-D echocardiography.
• To address the challenges posed by fuzzy boundaries and low image quality in echocardiographic segmentation.
• To enhance the discrimination of boundary pixels during the segmentation process.

Main Methods:

• Proposed a novel training strategy named multiconstrained aggregate learning (MCAL) that leverages anatomical knowledge from ground-truth labels.
• Incorporated a boundary distance transform weight (BDTW) into training objectives to emphasize boundary regions and improve segmentation accuracy.
• Developed an end-to-end framework with a top-down, bottom-up architecture featuring skip convolution fusion blocks.

Main Results:

• The MCAL strategy effectively utilizes anatomical knowledge to infer segmented parts and discriminate boundary pixels.
• The inclusion of BDTW significantly improved segmentation accuracy by enforcing higher weights on boundary regions.
• The proposed method demonstrated superior performance compared to other baseline segmentation models on two distinct datasets.

Conclusions:

• The multiconstrained aggregate learning (MCAL) framework offers a significant advancement in 2-D echocardiography segmentation.
• The method enhances the model's focus on anatomically relevant features and improves boundary pixel discrimination.
• MCAL provides a robust and accurate solution for left ventricular myocardium segmentation, benefiting clinical decision-making.