Fatemah-Noushin Golabchi

Automatic characterization of spinal cord histology images for comparisons to MR diffusion tensor measurements

Date: Fri 13 August

Abstract

Recently there has been a lot of interest in the study of nerve fiber tracts by means of Magnetic Resonance Imaging (MRI). Diffusion Tensor Magnetic Resonance Imaging (DT-MRI) techniques are believed to provide 3-dimensional qualitative and quantitative in-vivo information about the organization of the nerve fibers by measuring the 3-dimensional random motion of the water molecules within the tissue. The variation of the diffusion along different spatial directions in the tissue provides information about the diffusion anisotropy, and ultimately the tissue structure, non-invasively. Line Scan Diffusion tensor Imaging (LSDI) is one such MR technique, which has been developed recently. In order to validate LSDI generated DT-MRI data, there is a need to correlate these results with corresponding images from histological preparations. This correlation so far has been done manually and the results have been validated visually.

To automate this process we developed techniques for spatial domain processing on the histology images in an attempt to measure the same information DT-MRI techniques provides: orientation, anisotropy and density of the nerve fibers. This problem is characterized by a tremendous resolution mismatch (on the order of 1,000,000:1) between the histology and the DT-MRI images, and by the fact that the features of interest in the histology images are on the order of a few pixels in size, as well as by the inherent mismatch between the features measured by each modality, e.g. DT-MRI data measures water diffusion whereas only the cellular structure is available in the histology images. In particular, we describe two separate algorithms that use methods including Fourier transforms, non-linear morphological operators, and existing segmentation algorithms, to automatically analyze the histology images. One algorithm measures the anisotropy and orientation of nerve fibers in the plane of the histology images, while the other estimates a density measure for the nerve fibers which are transverse to the image plane. We present the results of our algorithms on images obtained from four slices of spinal cord tissue, and offer initial comparisons, in the case of transverse (through-plane) axons, to corresponding LSDI anisotropy measurements of the same tissue.

Committee:
Dr. W. Scott Hoge (Brigham and Women's Hospital)
Prof. Gilead Tadmor Prof. Dana Brooks (Advisor)