Threshold Relaxation is an Effective Means to Connect Gaps in 3 D Images of Complex Microvascular Networks

All optical histology (AOH) uses femtosecond pulse plasma mediated laser ablation in conjunction with two-photon laser scanning microscopy (TPLSM) to produce large anatomical volumes at micrometer-scale resolution. Specifically, we use AOH to produce ~1mm 3 datasets of cerebral vasculature with the goal of modeling its structural and physiological relationship to neuronal cells. Generating a binary mask of the cerebral vasculature is a first step towards this goal, and many methods have been proposed to segment such 3D structures. However, many analyses of the tubular vascular network (e.g., average vessel segment length, radii, point-to-point resistance and cycle statistics) are more efficiently computed on a vectorized representation of the data, i.e. a graph of connected centerline points. Generating such a graph requires sophisticated upstream algorithms for both segmentation and vectorization. Occasionally, the algorithms form erroneous gaps in the vectorized graph that do not properly represent the underlying anatomy. We present here a method to connect such gaps via local threshold relaxation. The method A) fills gaps by relaxing a binarization threshold on the grayscale data volume in the vicinity of each gap (found using the vectorization), B) computes a " bridging " strand for each gap, and C) produces a confidence metric for each " bridging strand ". We show reconnection results using our method on real 3D microvasculature data from the rodent brain and compare to a tensor voting method.