3D analysis of dissection photographs with surface scanning and machine learning for quantitative neuropathology.

When donated human brains are dissected at brain banks, they are first cut into slabs that are routinely photographed just with archiving purposes. Here we present a set of novel, open-source computational tools that enable 3D reconstruction of these photographs into a volumetric dataset, as well as subsequent 3D segmentation of the reconstructed volumes (i.e., assigning neuroanatomical labels to every pixel in the images). Our tools enable quantitative analyses of the brain at the macroscopic level (e.g., measuring volumes of brain regions) without requiring MRI scans of the specimens. The tools also provide a spatial mapping between an MRI (if available) and the photographs, thus enabling the study of correlations between macroscopic and microscopic features (e.g., atrophy vs histology).