A novel aortic valve segmentation from ultrasound image using continuous max-flow approach

Area of Science:

• Medical Imaging
• Cardiovascular Ultrasound
• Image Segmentation

Background:

• Accurate delineation of aortic valve geometric features is crucial for cardiac diagnostics, modeling, and interventions.
• Existing methods for segmenting aortic valves from ultrasound (US) images are not sufficiently addressed.

Purpose of the Study:

• To propose a novel algorithm for segmenting the aortic valve from intra-operative 2D short-axis US images.
• To achieve accurate and real-time aortic valve segmentation for improved clinical applications.

Main Methods:

• The algorithm employs probability estimation (composite probability estimation and single probability estimation) and a continuous max-flow (CMF) approach.
• It utilizes intensity and distance to the centroid from prior images to construct an energy function.
• Graphic Processing Unit (GPU) acceleration enables near real-time contour detection.

Main Results:

• Quantitative evaluation on 270 images from 3 subjects showed strong correlation with expert manual segmentation.
• Key metrics achieved: Average Symmetric Contour Distance (ASCD) of 1.79±0.46 pixels, Dice Metric (DM) of 0.96±0.01, and Reliability of 0.84.
• The algorithm processed images in approximately 39.23±5.02 ms per frame.

Conclusions:

• The developed algorithm provides accurate and efficient aortic valve segmentation from intra-operative ultrasound images.
• The real-time performance and high accuracy support its potential use in image-guided cardiac interventions.