DeepUS-ReconSeg: A multi-angle paired B-mode Ultrasound dataset for medical imaging reconstruction and segmentation

Area of Science:

• Medical Imaging
• Biomedical Engineering
• Computer Vision

Background:

• Deep learning models require large, high-quality datasets for training.
• Ultrasound imaging is a crucial diagnostic tool, but image quality can be a limitation.
• Accurate segmentation of anatomical structures in ultrasound is vital for diagnosis.

Purpose of the Study:

• To present a curated dataset of human forearm B-mode ultrasound images.
• To facilitate the development of deep learning algorithms for ultrasound image reconstruction and segmentation.
• To provide a publicly accessible resource for the medical imaging research community.

Main Methods:

• Collected 4,200 grayscale B-mode ultrasound images using a Verasonics Vantage 64LE system with an L11-5v transducer.
• Acquired images from 14 healthy subjects, capturing multiple anatomical views and orientations.
• Stored data in MATLAB .mat format and made it publicly available via Mendeley Data.

Main Results:

• A comprehensive dataset of 4,200 forearm ultrasound images has been successfully curated.
• The dataset includes high-resolution B-mode images suitable for advanced analysis.
• Data covers diverse anatomical views and orientations for robust model training.

Conclusions:

• The presented dataset is a valuable resource for advancing deep learning in medical image analysis.
• It will aid researchers in developing improved ultrasound image reconstruction and segmentation techniques.
• Public availability promotes collaborative research and accelerates innovation in ultrasound diagnostics.