Liver Ultrasound Image Quality Assessment for Deep Learning Enhancement

A study from 2026 by Jaeyoung Huh presents a dataset for assessing a deep learning algorithm's ability to enhance liver ultrasound images. The algorithm, based on a switchable cycle generative adversarial network (CycleGAN), was trained to improve images from a 12-year-old ultrasound device using images from a 4-year-old device as targets. Image quality metrics, including brightness, contrast, and overall quality, were evaluated by two experienced reviewers.

Use Cases

• Training image enhancement models based on paired low-quality and high-quality ultrasound images.
• Evaluating deep learning algorithms for improving diagnostic image quality based on reviewer assessments.
• Studying the impact of ultrasound device age on image quality based on the described device comparison.

Strengths

• Dataset is openly licensed under CC-BY-4.0.
• Image quality was assessed by two experienced reviewers, with inter-reader agreement metrics (weighted kappa values ranging from 0.225 to 0.838) provided.
• The deep learning approach demonstrated statistically significant improvements in brightness, contrast, and overall quality (p < 0.001).

Limitations

• Row count is unknown, which may limit suitability assessment.
• Column-level documentation is absent; field semantics must be inferred after download.
• The dataset is very small at 5.5 KB, indicating limited scope, likely containing summary metrics rather than raw images.

Provenance

Source

figshare

Collection Method

Consecutively acquired grey-scale liver ultrasound examinations from two ultrasound devices.

Time Range

2026

Freshness

Last updated 2026-04-28 17:44:23; freshness should be verified.

Geography

null