Authors: Alex Ranne, Liming Kuang, Yordanka Velikova, Nassir Navab, Ferdinando Rodriguez y Baena
Published on: March 21, 2024
Impact Score: 7.8
Arxiv code: Arxiv:2403.14465
Summary
- What is new: A self-supervised deep learning architecture using AiAReSeg with Attention in Attention mechanism for segmenting catheters in longitudinal ultrasound images without labeled data.
- Why this is important: Minimally invasive endovascular procedures rely on contrast-enhanced angiography, posing health risks due to radiation exposure. Ultrasound could be an alternative but is challenging to interpret.
- What the research proposes: The proposed solution is a novel deep learning model that segments catheters in ultrasound images without requiring labeled data, utilizing synthetic data and a unique CT-Ultrasound domain translation.
- Results: The model showed promising results in segmenting catheters in both synthetic data and images collected from silicon aorta phantoms, indicating potential for future clinical application.
Technical Details
Technological frameworks used: AiAReSeg, a segmentation transformer with Attention in Attention mechanism. CT-Ultrasound common domain (CACTUSS) for data translation. FlowNet2 for optical flow computation.
Models used: nan
Data used: Synthetic ultrasound data based on physics-driven simulations and real images from silicon aorta phantoms.
Potential Impact
Radiology and medical imaging companies, healthcare providers performing endovascular procedures. Potential benefits for training programs in interventional radiology.
Want to implement this idea in a business?
We have generated a startup concept here: EchoNavigate.
Leave a Reply