Aligning image analysis tools with clinical practice in multiple sclerosis

In this thesis, research was conducted to improve our understanding of MRI-based measures for monitoring and managing multiple sclerosis (MS). The aim was to identify solutions to overcome current challenges in MS imaging, by evaluating the reliability, feasibility, and clinical applicability of automated imaging metrics. Through these efforts, the goal is to improve the use and adoptation of (advanced) imaging tools in clinical practice, ultimately enhancing patient care. An important milestone was the acquisition of the AMS2 dataset, which facilitated a large part of the thesis and can continue to reliability research. This thesis comprises multiple studies focused on the reliability of existing MS imaging measures, optimizing their clinical application, and tailoring them for improved disease management. Chapter 2.1 evaluated brain atrophy measurements across three MRI scanners and six software packages, revealing high within-scanner repeatability but considerable between-scanner variability, underscoring the need for scanner and software consistency. Chapter 2.2 focused on brain-predicted age difference (brain-PAD) and found excellent within-scanner repeatability but inconsistent reproducibility between scanners and models, highlighting technical limitations for clinical adoption. Chapter 3.1 demonstrated that scanner-specific optimization of automated lesion segmentation tools significantly improved accuracy and reproducibility, addressing a major barrier to implementation. Chapter 3.2 showed that brain volume measurements from 3D-FLAIR sequences were comparable to those from 3D T1-weighted images and correlated similarly with clinical outcomes, supporting their use in routine care. In Chapter 4.1 we developed MS-specific brain volume reference curves that better correlated with disease severity and cognitive decline than healthy control-based models, improving clinical relevance. Chapter 4.2 identified key clinician requirements for quantitative radiology reports (QReports), such as lesion and brain volume metrics and workflow integration, while noting concerns about validation, cost, and report generation time. This thesis highlights the need for improving the reliability and clinical alignment of MRI-based imaging tools in MS care, showing that both technical optimization and clinician-centered design are essential for successful implementation. Future research should focus on validating these tools in larger clinical settings and assessing their long-term impact on disease monitoring and treatment decision-making.