Computer Graphics and Visualization • Publications

Computer Graphics Forum

Visualization of 4D Blood-Flow Fields by Spatiotemporal Hierarchical Clustering

R.F.P. van Pelt, S.S.A.M. Jacobs, B.M. ter Haar Romeny, and Anna Vilanova

Overview of the visualization framework, based on spatiotemporal hierarchical clustering. The gray dashed arrows depict pre-processing steps. (1) A tMIP volume is generated, and (2) an iso-threshold captures the voxels that are clustered. (3) Next, the cluster hierarchy is constructed. (4) Using the cluster tree, labels are generated per cardiac phase. After preprocessing, the real-time visualization is generated using the available data structures, as depicted by the solid blue arrow.

Advancements in the acquisition and modeling of flow fields result in unsteady volumetric flow fields of unprecedented quality. An important example is found in the analysis of unsteady blood-flow data. Preclinical research strives for a better understanding of correlations between the hemodynamics and the progression of cardiovascular diseases. Modern-day computer models and MRI acquisition provide time-resolved volumetric blood-flow velocity fields. Unfortunately, these fields often remain unexplored, as high-dimensional data are difficult to conceive. We present a spatiotemporal, i.e., four-dimensional, hierarchical clustering, yielding a sparse representation of the velocity data. The clustering results underpin an illustrative visualization approach, facilitating visual analysis. The hierarchy allows an intuitive level-of-detail selection, largely retaining important flow patterns. The clustering employs dissimilarity measures to construct the hierarchy. We have adapted two existing measures for steady vector fields for use in the spacetime domain. Because of the inherent computational complexity of the multidimensional clustering, we introduce a coarse hierarchical clustering approach, which closely approximates the full hierarchy generation, and considerably improves the performance. The resulting clusters are visualized by representative patharrows, in combination with an illustrative anatomical context. We present various seeding approaches and visualization styles, providing sparse overviews of the unsteady behavior of volumetric flow fields.

More Information

Publisher Paper preprint

Citation

R.F.P. van Pelt, S.S.A.M. Jacobs, B.M. ter Haar Romeny, and Anna Vilanova, Visualization of 4D Blood-Flow Fields by Spatiotemporal Hierarchical Clustering, Computer Graphics Forum, 31, pp. 1065–1074, 2012.

BibTex

@article{bib:van pelt:2012,
    author       = { van Pelt, R.F.P. and Jacobs, S.S.A.M. and ter Haar Romeny, B.M. and Vilanova, Anna },    
    title        = { Visualization of 4D Blood-Flow Fields by Spatiotemporal Hierarchical Clustering },
    journal      = { Computer Graphics Forum },
    volume       = { 31 },
    year         = { 2012 },
    pages        = { 1065--1074 },
    doi          = { 10.1111/j.1467-8659.2012.03099.x },
    dblp         = { journals/cgf/PeltJRV12 },
    url          = { https://publications.graphics.tudelft.nl/papers/317 },
}