Paired Point-of-Care Ultrasound Dataset for Image Quality Enhancement

A dataset of 1,064 paired ultrasound image sets from ex vivo human tissue and a phantom, collected using a custom motorized gantry for spatial alignment. Created by Lennard van Karnenbeek and published in 2026, it contains 2,128 total PNG images preprocessed for consistency. The data is intended for research on ultrasound image quality enhancement using deep learning methods.

Use Cases

• Training conditional GANs for ultrasound image enhancement based on the paired low-quality and high-quality images.
• Benchmarking super-resolution models using the spatially aligned image pairs from ex vivo tissue and phantoms.
• Developing models for translating point-of-care ultrasound images to cart-based ultrasound quality based on the device-paired data.

Strengths

• Contains 1,064 paired image sets (2,128 total images) with true spatial alignment ensured by a custom motorized gantry.
• Includes data from 15 ex vivo human specimens (breast, sarcoma, colorectal) and an abdominal phantom, providing a mix of real tissue and controlled phantom data.
• Images are preprocessed (cropped, resampled, intensity normalized) and fully anonymized for research use.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Data is derived from ex vivo specimens and a phantom, which may not fully represent in vivo clinical imaging conditions.

Provenance

Source

figshare, author Lennard van Karnenbeek.

Collection Method

Acquired using a custom motorized gantry setup to scan ex vivo tissue specimens and phantoms at matching spatial coordinates, with additional phantom-based registration.

Freshness

Last updated 2026-05-19 08:11:52