Progressive extraction of neural models from high-resolution 3d images of brain

In the last few years, the fast progresses of scanning technologies have produced a great increment of resolution of 2D/3D images, measured by the number of distinct pixels in each dimension that can be displayed on a display device. This growth of resolution is about one order of magnitude (10 times) in each single spatial dimension, and therefore of second order in the number of pixels in 2D and of order three for the number of 3D voxels. Hence, the sectional image of a neuronal or vessel structure in the brain, depicted in the past --- at maximum resolution --- by few pixels in its average section, say less than 10x10, is currently reproduced by the newest imaging technologies as measuring more than ten times that number.Conversely, in this research we aim to experiment a novel decompositive approach [3] to the computation of boundaries of biological structures to be extracted from extreme-resolution three- dimensional images. The new method is based on the topological extraction of boundaries of “chains” of image elements, as performed by boundary and coboundary maps between the linear spaces generated by the cell decomposition of the image, possibly followed by a non-iterative Laplacian smoothing of the extracted surface, in order to smooth-out the resulting model.