Medical Volume Visualization Beyond Single Voxel Values

• Main Author: Lindholm, Stefan
• Format: Doctoral Thesis
• Language: English
• Published: Linköpings universitet, Medie- och Informationsteknik 2014
• Online Access: http://urn.kb.se/resolve?urn=urn:nbn:se:liu:diva-110239; http://nbn-resolving.de/urn:isbn:978-91-7519-256-7 (print)

Medical visualization involves many complex decisions for both the user and the imaging algorithms. This thesis aims to improve medical volume visualization through a series of technical contributions to aid such decision processes. Improvements are achieved by using more data, beyond single voxels, in the associated visual analyses. Simultaneous visualization of multiple data sources and different data formats is rapidly becoming a necessity. This is due to both the growing number of data producing image acquisition techniques as well as the increase in geometric data representations that can be created. Maintaining high rendering performance under these circumstances is challenging, but necessary, to support an exploratory visualization process. This thesis proposes two algorithms to address this challenge: a multi-volume approach that applies binary-space partitioning to solve painters' algorithm geometrically and a rendering algorithm for hybrid data that improves the management of the available graphics memory. Additional information for decision support is often derived from the captured image data. Classification techniques, in particular, often utilize secondary information sources or neighborhood analysis as means to improve specificity. One example is a proposed algorithm that improves visualization of blood vessels by automatically optimizing visualization parameters based on observed vesselness. This thesis also proposes algorithms involving neighborhood analysis, with a particular focus on domain specific classification knowledge provided by the user. One algorithm provides the ability to semantically state spatial relations between tissues based on encoded material information. Another algorithm improves the representation of discrete features by integrating the users' knowledge in the reconstruction step of the visualization pipeline. Many of the methods proposed in this thesis can also be applied to other domains, but are all described here in the context of medical volume visualization as most of the research has been performed within this field.