Whole-Mouse Brain Vascular Analysis Framework: Synthetic Model-Based Validation, Informatics Platform, and Queryable Database

The past decade has seen innovative advancements in light microscopy instrumentation that have afforded the acquisition of whole-brain datasets at micrometer resolution. As the hardware and software used to automate the traditional neuroanatomical workflow become more accessible to researchers around the globe, so will the tools needed to analyze whole-brain datasets. Only recently has the focus begun to shift from the development of instrumentation towards platforms for data-driven quantitative analyses. As a consequence of this, the tools required for large-scale quantitative studies across the whole brain are few and far between. In this dissertation, we aim to change this through the development of a standardized, quantitative approach to the study of whole-brain, cerebrovasculature datasets. Our standardized and quantitative approach has four components. The first is the construction of synthetic cerebrovasculature models that can be used in conjunction with the second component, a model-based validation system. Any cerebrovasculature study conducted using imaging data must first extract the filaments embedded within that dataset. The segmentation algorithms that are commonly used to do this are frequently validated on small-scale datasets that represent only a small selection of cerebrovasculature variability. The question is how do these algorithms perform when applied to large-scale datasets. Our model-based validation system uses biologically inspired, large-scale datasets that asses the accuracy of the segmentation algorithm output against ground truth data. Once the data is segmented, we have implemented an informatics platform that calculates descriptive statistics across the entire volume. Attributes describing each vascular filament are also calculated. These include measures of vascular radius, length, surface area, volume, tortuosity, and others. The result is a massive amount of data for the cerebrovasculature segments. The question becomes how can this be analyzed sensibly. Given that both cerebrovasculature topology and geometry can be capture in graph form, we construct the fourth component of our system: a graph database that stores the cerebrovasculature. The graph model of cerebrovasculature that we have developed allows segments to be searched across the whole-brain based on their attributes and/or location. We also implemented a means to reconstruct the segments returned by a specific query for visualizations. This means that a simple text-based query can retrieve cerebrovasculature geometry and topology of the specified vasculature. For example, a query can return all vessels within the frontal cortex, those with specific attribute(s) value range(s), or any combination of attribute and location. Complex graph algorithms can also be applied, such as the shortest path between two bifurcation points or measures of centrality that are important in determining the robust and fragile aspects of blood flow through the cerebrovasculature system. To illustrate the utility of our system, we construct a whole-brain database of vascular connectivity from the Knife-Edge Scanning Microscope India Ink dataset. Using our cerebrovasculature database, we were able to study the cerebrovasculature system by issuing text-based queries to extract the vessel segments that we were interested in. The outcome of our investigation was a wealth of information about the cerebrovasculature system as a whole, and about the different classifications of vessels comprising it. The results returned from these simple queries even generated some interesting and biologically relevant questions. For instance, the profound spikes in radius distribution for some classes of vessels that did not present in other classes. We expect that the methods described in this dissertation will open the door for data-driven, quantitative investigation across the whole-brain. At the time of writing – and to the best of our knowledge that prior to this work – there was not a systemic way to assess segmentation algorithm performance, calculate attributes for each segment of vasculature extracted across the whole brain, and store those results in a queryable database that also stores geometry and topology of the entire cerebrovasculature system. We believe that our method can and will set the standard for largescale cerebrovasculature research. Therefore, in conclusion, we state that our methods contribute a standardized, quantitative approach to the study of cerebrovasculature datasets acquired using modern imaging techniques