[An attention-guided network for bilateral ventricular segmentation in pediatric echocardiography]

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Cardiology

Context:

• Pediatric echocardiograms present unique segmentation challenges due to rapid heart size changes and faster heart rates, leading to blurred boundaries.
• Accurate segmentation of cardiac structures is crucial for diagnosing congenital heart disease in children.

Purpose:

• To develop and evaluate a novel dual decoder network model incorporating channel and scale attention for accurate pediatric echocardiogram segmentation.
• To improve the delineation of left and right ventricular regions and their boundaries in pediatric cardiac ultrasound images.

Summary:

• A dual decoder network model with attention mechanisms (channel and scale attention) and deep supervision was proposed for pediatric echocardiogram segmentation.
• The model utilizes an attention-guided decoder and skip connections to enhance ventricular feature representation and edge accuracy.
• The proposed method achieved an average Dice coefficient of 90.63% on a bilateral ventricular segmentation dataset.

Impact:

• The developed method offers a significant improvement over existing techniques for pediatric echocardiographic bilateral ventricular segmentation.
• Enhanced accuracy in ventricular segmentation, particularly at the edges, can aid in the auxiliary diagnosis of congenital heart disease.
• This research provides a new computational solution for analyzing pediatric cardiac ultrasound images, potentially improving diagnostic workflows.