Semi-Quantitative Analysis of Magnetic Resonance Imaging in Arthritis: The Pursuit of Optimal Scoring Granularity
Date
Author
Institution
Degree Level
Degree
Department
Supervisor / Co-Supervisor and Their Department(s)
Citation for Previous Publication
Link to Related Item
Abstract
Semi-quantitative image scoring systems have been developed to assess various forms of arthritis through assignment of numerical scores corresponding to the extent of lesion presence within an image. Such scoring methodologies allow researchers to systematically record differences between patients and within the same patient over time, allowing for detailed analyses of imaging findings as they relate to disease severity, treatment response, and various clinical markers. The level of detail, or granularity, contained within a scoring system influences the ability of the score to convey smaller differences or changes in lesion extent between images, but can also impact the reliability of scores assigned to the same images by different readers as well as the amount of time and effort required of readers. This thesis examines how scoring granularity impacts the utility of semi-quantitative image data, as well as inter-rater reliability.
Chapter 1 contains a study of inflammatory and structural lesion scoring in the sacroiliac joint of ankylosing spondylitis patients at different levels of granularity, showing how analysis of semi-quantitative scoring data can produce different results depending on which reader’s data is used.
Chapters 2 and 3 both deal in semi-quantitative scoring of bone marrow lesions in knee osteoarthritis patients. Chapter 2 demonstrates how reliability
of scoring can–but does not always– decrease as the regions of interest decrease in size, while chapter 3 covers an extensive analysis of bone marrow lesions at different levels of detailed scoring, and explores the use of an artificial intelligence-generated semi-quantitative scoring output to test these results in a larger dataset. This final chapter suggests a justification for scoring in granular detail, and makes a case for the need to ensure training of artificial intelligence algorithms is targeted for reliable detailed scoring.