Low-dose computed tomography perceptual image quality assessment.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Radiology

Background:

• Optimizing radiation dose and image quality in computed tomography (CT) is vital for patient safety.
• Radiologist assessments are gold standard but time-consuming; objective metrics like PSNR and SSIM have limitations for CT.
• Developing deep learning (DL) based image quality assessment (IQA) aligned with human perception is a growing need.

Purpose of the Study:

• To address the lack of open-source datasets and benchmark models for CT IQA.
• To facilitate the development of DL models for perceptual CT image quality assessment.
• To introduce a novel dataset and benchmark for evaluating CT IQA methods.

Main Methods:

• Organized the Low-dose Computed Tomography Perceptual Image Quality Assessment Challenge (MICCAI 2023).
• Created the first open-source CT IQA dataset with 1,000 images and radiologist scores.
• Benchmarked six submitted DL-based IQA methods.

Main Results:

• The challenge introduced a valuable, open-source dataset for CT IQA research.
• Analysis of six methods provided insights into current CT IQA performance.
• Demonstrated the potential of no-reference IQA methods to outperform full-reference methods.

Conclusions:

• The novel CT IQA dataset is a significant contribution to the research community.
• This work paves the way for advanced, perception-aligned CT image quality assessment tools.
• Future development of no-reference IQA methods holds promise for exceeding current full-reference capabilities.