Real-time aortic valve segmentation from transesophageal echocardiography sequence

Area of Science:

• Medical Imaging
• Cardiovascular Imaging
• Image Segmentation

Background:

• Accurate geometric features of the aortic valve are vital for clinical diagnostics, shape modeling, and image-guided cardiac interventions, particularly transcatheter aortic valve implantation (TAVI).
• Segmentation of the aortic valve from transesophageal echocardiography (TEE) sequences is challenging, with limited existing research.

Purpose of the Study:

• To present a novel real-time method for segmenting the aortic valve from intraoperative, short-axis view TEE sequences.
• To provide accurate segmentation to support TAVI procedures.

Main Methods:

• The method employs a two-stage approach: improved probability estimation (composite and single probability estimation) using prior frames and energy function construction.
• A continuous max-flow (CMF) algorithm, accelerated by a GPU, is utilized for real-time contour extraction.

Main Results:

• The algorithm demonstrated high accuracy in segmenting the aortic valve, with an average symmetric contour distance (ASCD) of 0.85 ± 0.21 mm and a Dice metric (DM) of 0.96 ± 0.01.
• Real-time performance was achieved, with a computation time of 57.04 ± 8.98 ms per frame, and high reliability (0.90).

Conclusions:

• The proposed method accurately segments the aortic valve in real-time from short-axis TEE sequences.
• This technique offers valuable support for clinical applications like TAVI.