| dc.description.abstract | The kidney, a vital organ responsible for waste excretion and hormone regulation, is frequently afflicted by non-neoplastic diseases affecting components like glomeruli and tubules. Chronic kidney disease (CKD), often progressing to end-stage kidney disease (ESKD), presents a substantial global health challenge. With nearly 100 million Europeans affected by CKD and projections indicating it may become a leading cause by 2040, there is an urgent need for advanced diagnostics, particularly in the under-researched area of non-neoplastic kidney diseases.
Digital pathology (DP) has transformed biomedical research and pathology diagnostics through the utilization of Whole Slide Images (WSIs) for precise tissue analysis. Image registration, a key application in DP, enables the alignment of histological sections, facilitating accurate comparisons and comprehensive tissue analysis. This thesis endeavours to integrate image registration tool into a future pipeline designed to automate the alignment and ordering of non-neoplastic kidney biopsy sections.
The thesis begins with a literature review to find image registration tools. Then the image registration tools found; HistokatFusion (commercial) and TIAToolbox (open source) were evaluated for a potential integration into a future pipeline aimed at automating the alignment and ordering of non-neoplastic kidney biopsy sections from Haukeland University Hospital, bergen. Through the application of Intersection Over Union (IOU) and Dice Score (DSC) metrics, the effectiveness of these tools is assessed, leading to the selection of TIAToolbox for further analysis due to its open-source nature.
Parameters such as grid spacing and sampling percentages were evaluated for non-rigid registration using TIAToolbox for alignment quality and computational efficiency. Grid spacing 200 and sampling percentage 1.0 show promising results and were used for further investigation into diverse cases which showed successful registration. The thesis also addresses specific challenges encountered in three difficult cases, highlighting the need for continued refinement of the pre-processing pipeline to accommodate complex or damaged tissue sections.
Areas for improvement include automating the pre-processing pipeline, evaluating additional registration tools, and exploring methods beyond pixel-based metrics like IOU and DSC. Further testing on a broader range of WSIs is necessary to comprehensively assess the pipeline’s registration success. Future research should address these limitations and integrate the pipeline into a larger framework for the automatic alignment and ordering of biopsy sections, enhancing efficiency for nephropathologists. Overall, this study provides valuable insights into image registration for non-neoplastic kidney biopsy analysis. | |
