Spatial description-based approach towards integration of biomedical atlases

Abstract

Biomedical imaging has become ubiquitous in both basic research and the clinical sciences. As technology advances the resulting multitude of imaging modalities has led to a sharp rise in the quantity and quality of such images. Whether for epi- demiological studies, educational uses, clinical monitoring, or translational science purposes, the ability to integrate and compare such image-based data has become in- creasingly critical in the life sciences and eHealth domain. Ontology-based solutions often lack spatial precision. Image processing-based solutions may have di culties when the underlying morphologies are too di erent. This thesis proposes a compro- mise solution which captures location in biomedical images via spatial descriptions. Three approaches of spatial descriptions have been explored. These include: (1) spatial descriptions based on spatial relationships between segmented regions; (2) spatial descriptions based on ducial points and a set of spatial relations; and (3) spatial descriptions based on ducial points and a set of spatial relations, integrated with spatial relations between segmented regions. Evaluation, particularly in the context of mouse gene expression data, a good representative of spatio-temporal bi- ological data, suggests that the spatial description-based solution can provide good spatial precision. This dissertation discusses the need for biomedical image data in- tegration, the shortcomings of existing solutions and proposes new algorithms based on spatial descriptions of anatomical details in the image. Evaluation studies, par- ticularly in the context of gene expression data analysis, were carried out to study the performance of the new algorithms.