Visual Analysis in Population Imaging Research The VAnPIRe project aims at the development and evaluation a new interactive visual analysis approach that enables the extraction of patterns and high-potential hypotheses, from the irregular, multi-timepoint, mixed imaging, genetics and other diagnostic data acquired in population imaging research. New insights into the working of the clinical outcome of a population can be effectively extracted from a great deal of heterogeneous study data by augmenting the human visual system and cognition with interactive visualization and coupled feature extraction techniques.
We developed an approximated version of the tSNE dimensionality reduction algorithm (A-tSNE). We demonstrated that the computation time can be reduced by two orders of magnitude by using our approximated algorithm for the computation of high-dimensional similarities. We developed a visualization system that allows the inspection of the degree of approximation in the embedding as well as steering the refinement of the approximation.
We created an integrated visual analysis system that allows visualization of and interaction with high dimensional single cell data at multiple scales. The integrated nature of the system, in combination with techniques developed previously within this project, such as A-tSNE, allows rapid analysis of the data.
We develop HSNE, a hierarchical dimensionality reduction algorithm. The algorithm provides hierarchical embedding and exploration, allowing the analysis of datasets consisting of millions of points, at interactive rates. At every scale, the algorithm reveals clusters of points of increasing size. Compared to other hierarchical embedding algorithms HSNE retains non-linear relationships throughout the hierarchy, while it is much faster and enables exploration of larger data sets compared to other non-linear (non-hierarchical) embedding techniques.