A computational method to quantitate neuronal cell loss in a transgenic mouse model of Alzheimer's disease

Abstract

Alzheimer's disease (AD) neuropathology is characterized by key features that include the deposition of the amyloid-beta peptides into senile plaques, the formation of neurofibrillary tangles, and the loss of neurons and synapses in specific brain regions. Using MR microscopy, we have detected loss of neurons in the pyramidal cell layer of the hippocampal CA1 subfield (PLCA1) in 5xFAD transgenic mice, a very rapid progression AD model. Here we report a novel semi-automatic PLCA1 segmentation method based on unsupervised support vector machines (SVM), which uses the image voxels distance to the classification hyperplane to quantify this age dependent neuronal loss. RARE 3D-images of fixed brains were acquired on a 14.1T microimager using TR/TEeff 2500ms/40ms and pixel size 35μm x35μm x35μm. Voxels are described using two 3D “sheetness” features that match the PLCA1 MR appearance. Laplacian L(I) = div (grad (I)) is used to model the neuronal layer core, defined as regions with small derivatives, surrounded by neighbors with rapidly increasing intensity. The second feature is the largest positive eigenvalue of the Hessian matrix Hjk(I) = DjDk(I), the square matrix of second-order partial derivatives. The distance to the SVM separation hyperplane (SVMDist) is used to measure the neuronal cell loss as indicated by the loss of MR contrast. The algorithm was applied to a manually delineated volume of interest around the PLCA1 of 2, 4, and 10 months old 5xFAD transgenic mice. MR image contrast, PLCA1 volume and average SVMDist decreased with age for the 5xFAD mice indicating CA1 neuronal loss. The WT mice showed increase in SVMDist and no change in the PLCA1 volume. Our measures of neuronal cell loss parallel the known evolution of amyloid deposition in 5xFAD mice which begins at 2 months of age and reaches a very large burden by 9-10 months of age. SVMDist distribution may be useful to understand AD cell loss mechanisms. Unlike senile plaques, neuronal cell loss shows strong correlation with cognitive decline in AD. Therefore, the presented technique could be used for tracking AD evolution and asses emerging AD therapies.