Abstract-A typical data-driven visualization of electroencephalography (EEG) coherence is a graph layout, with vertices representing electrodes and edges representing significant coherences between electrode signals. A drawback of this layout is its visual clutter for multichannel EEG. To reduce clutter, we define a functional unit (FU) as a data-driven region of interest (ROI). An FU is a spatially connected set of electrodes recording pairwise significantly coherent signals, represented in the coherence graph by a spatially connected clique. Earlier we presented two methods to detect FUs, a maximal clique based (MCB) method (time complexity O(3 n/3 ), with n the number of vertices) and a more efficient watershed based (WB) method (time complexity O(n 2 log n)). To reduce the potential over-segmentation of the WB method, we introduce an improved watershed based (IWB) method (time complexity O(n 2 log n)). The IWB method merges basins representing FUs during the segmentation if they are spatially connected and if their union is a clique. The WB and IWB method both are up to a factor of 100,000 faster than the MCB method for a typical multichannel setting with 128 EEG channels, thus making interactive visualization of multichannel EEG coherence possible. Results show that, considering the MCB method as the gold standard, the difference between IWB and MCB FU maps is smaller than between WB and MCB FU maps. We also introduce two novel group maps for data-driven group analysis as extensions of the IWB method. First, the group mean coherence map preserves dominant features from a collection of individual FU maps. Second, the group FU size map visualizes the average FU size per electrode across a collection of individual FU maps. Finally, we employ an extensive case study to evaluate the IWB FU map and the two new group maps for data-driven group analysis. Results, in accordance with conventional findings, indicate differences in EEG coherence between younger and older adults. However, they also suggest that an initial selection of hypothesis-driven ROIs could be extended with additional datadriven ROIs.

Decision trees have been shown to be effective at classifying subjects with Parkinson's disease when provided with features (subject scores) derived from FDG-PET data. Such subject scores have strong discriminative power but are not intuitive to understand. We therefore augment each decision node with thumbnails of the principal component (PC) images from which the subject scores are computed, and also provide labeled scatter plots of the distribution of scores. These plots allow the progress of individual subjects to be traced through the tree and enable the user to focus on complex or unexpected classifications. In addition, we present a visual representation of a typical brain activity pattern arriving at each leaf node, and show how this can be compared to a known reference to validate the behaviour of the tree.