Modelling visual perception and image quality for medical imaging

Ma, Yueran 2025. Modelling visual perception and image quality for medical imaging. PhD Thesis, Cardiff University. Item availability restricted.

	PDF - Accepted Post-Print Version Restricted to Repository staff only until 19 May 2027 due to copyright restrictions. Available under License Creative Commons Attribution Non-commercial No Derivatives. Download (26MB) | Request a copy
	PDF (Cardiff University Electronic Publication Form) - Supplemental Material Restricted to Repository staff only Download (467kB) | Request a copy

Abstract

Image quality is a critical determinant of diagnostic accuracy in medical imaging, influencing both radiologists’ confidence and their ability to detect clinically relevant features. This thesis investigates computational modelling of medical image quality perception and radiologists’ visual search patterns, with the overarching aim of bridging image quality assessment (IQA) methodologies and human perceptual processes in clinical contexts. First, a benchmark MRI IQA database, RAD-IQMRI, was constructed using simulated artifacts and annotated through controlled subjective experiments with experienced radiologists, providing a reliable resource for evaluating IQA methods under standardised conditions. Analysis of this dataset revealed the influence of clinical experience on perceptual judgements, showing consistent intra-group agreement and differences in quality sensitivity across expertise levels. Building on this foundation, a novel dual-branch deep learning framework, MIQANet, was developed to predict MRI image quality scores by jointly modelling global structural information and local fine details. Extensive evaluations demonstrated that MIQANet outperformed state-of-the-art IQA models across multiple datasets, with superior generalisation capability. The research then expanded to the analysis of radiologists’ visual search patterns using a newly collected chest X-ray eye-tracking database. Statistical modelling showed that experienced radiologists achieved higher diagnostic accuracy with fewer fixations and shorter total viewing times, while less experienced readers employed more exhaustive search strategies. Furthermore, task-dependent adaptations were observed, with distinct scanpath characteristics when interpreting malignant versus non-malignant cases. Collectively, these studies advance understanding of how image quality perception and search behaviour interact with diagnostic performance, and establish methodological and computational tools for developing IQA systems that are both technically robust and clinically informed. By integrating controlled perceptual experiments, deep learning-based IQA modelling, and eye-tracking analysis, this work provides a foundation for enhancing image interpretation efficiency and reliability in medical imaging practice.

Item Type:	Thesis (PhD)
Date Type:	Completion
Status:	Unpublished
Schools:	Schools > Computer Science & Informatics
Subjects:	Q Science > QA Mathematics > QA75 Electronic computers. Computer science
Date of First Compliant Deposit:	19 May 2026
Date of Acceptance:	18 May 2026
Last Modified:	22 May 2026 09:37
URI:	https://orca.cardiff.ac.uk/id/eprint/187113

Actions (repository staff only)

Edit Item