Deep convolutional neural networks for segmenting 3D in vivo multiphoton images of vasculature in Alzheimer disease mouse models

摘要

The health and function of tissue rely on its vasculature network to providereliable blood perfusion. Volumetric imaging approaches, such as multiphotonmicroscopy, are able to generate detailed 3D images of blood vessels that couldcontribute to our understanding of the role of vascular structure in normalphysiology and in disease mechanisms. The segmentation of vessels, a core imageanalysis problem, is a bottleneck that has prevented the systematic comparisonof 3D vascular architecture across experimental populations. We explored theuse of convolutional neural networks to segment 3D vessels within volumetric invivo images acquired by multiphoton microscopy. We evaluated different networkarchitectures and machine learning techniques in the context of thissegmentation problem. We show that our optimized convolutional neural networkarchitecture, which we call DeepVess, yielded a segmentation accuracy that wasbetter than both the current state-of-the-art and a trained human annotator,while also being orders of magnitude faster. To explore the effects of agingand Alzheimer's disease on capillaries, we applied DeepVess to 3D images ofcortical blood vessels in young and old mouse models of Alzheimer's disease andwild type littermates. We found little difference in the distribution ofcapillary diameter or tortuosity between these groups, but did note a decreasein the number of longer capillary segments ($>75mu m$) in aged animals ascompared to young, in both wild type and Alzheimer's disease mouse models.