Speaker: Nicolas Swoboda (Inst. 193-02)
In their studies of the brain of the common fruit fly Drosophila melanogaster, neurobiologists investigate neural connectivity with the goal to discover how complex behaviour is generated. Gaining information on potential connectivity between neurons is an essential step in their workflow. This thesis presents a way to compute and visualise such potential connectivity information from segmented neurons. It shortens the tedious month-long search for potential connectivity down to a few minutes.
Overlaps of arborisations of two or more neurons indicate a potential anatomical connection, and thus a potential functional connection. The computation of this data starts from neuron meshes. The meshes—segmented from confocal light-microscopy images of the fruit fly—are intersected to find overlapping areas, i.e. areas of potential anatomical connectivity. This information can then help to discover actual functional connectivity in a neural circuit.
Analysing higher order overlaps, i.e., intersections of more than two arborisations of segmented neuron data in the same location, poses new challenges. The visualisation in 2D sections or 3D is impeded by visual clutter and occlusion. Computation of relevant volumetric information becomes difficult for higher order overlaps, because the number of possible overlaps increases exponentially with the number of arborisations. This makes the pre-computation for all possible combinations infeasible. Previous tools have thus been restricted to the quantification and visualisation of pairwise overlaps.
The thesis presents a novel solution addressing these issues for higher order overlaps. A novel abstracting design is coupled with a modern approach for on-demand GPU computations. Our tool calculates for the first time volumetric information of higher order overlaps on the GPU using A-buffers. The thesis addresses the visual complexity of the data with the implementation of an innovative novel design created by the graphics designer Judith Moosburner. We realised this design using non-photorealistic rendering techniques and perspicuous user interfaces, including interactive glyphs and linked views on quantitative overlap information. To complement the neuroscientists’ workflows the resulting interactive tool has been integrated into BrainGazer, a software tool for advanced visualisation and exploration of neural images and circuit data. Quantitative evaluation with neuroscientists and non-expert users demonstrated the utility and usability of the tool.