Ultrasound segmentation analysis via distinct and completed anatomical borders

1

Ultrasound imaging is crucial for non-invasive diagnostics, particularly in obstetrics, cardiology, and oncology.
2

Segmenting ultrasound images is challenging due to noise, artifacts, and low contrast, complicating anatomical boundary delineation.
3

The study introduces a novel evaluation approach for ultrasound segmentation networks, focusing on distinct and completed anatomical borders.
4

A new open-source LEG-3D-US dataset is provided to enhance research reproducibility and facilitate advancements in ultrasound segmentation.
5

Grad-CAM is utilized to analyze network activations, highlighting regions of interest in ultrasound images for improved model interpretability.

International Journal Of Computer Assisted Radiology And Surgery

by Vanessa Gonzalez Duque, Alexandra Marquardt, Yordanka Velikova, Lilian Lacourpaille, Antoine Nordez, Marion Crouzier, Hong Joo Lee, Diana Mateus, Nassir Navab
May 25, 2024

1