Methodology for computing white matter nerve fiber orientation in human histological slices

• Published in: Journal of neuroscience methods 2016-03, Vol.261, p.75-84
• Main Authors: Nazaran, Amin, Wisco, Jonathan J., Hageman, Nathan, Schettler, Stephen P., Wong, Anita, Vinters, Harry V., Teng, Chia-Chi, Bangerter, Neal K.
• Format: Article
• Language: English
• Subjects: Algorithms; Average orientation; Fiber tracking; Fourier Analysis; Fourier transform; Hippocampus - anatomy & histology; Histological Techniques - methods; Histology; Humans; Image Processing, Computer-Assisted - methods; Nerve fibers; Nerve fibers orientation; Nerve Fibers, Myelinated; Orientation map; Pattern Recognition, Automated - methods; White matter; White Matter - anatomy & histology
• ISSN: 0165-0270; 1872-678X
• DOI: 10.1016/j.jneumeth.2015.11.023

•We present a segmentation algorithm for histologically stained white matter fibers.•The algorithm extracts directionality of white matter fibers in the histological slices.•The algorithm was validated against manual segmentation by an expert anatomist.•The algorithm is accurate and reliable. The gold standard for mapping nerve fiber orientation in white matter of the human brain is histological analysis through biopsy. Such mappings are a crucial step in validating non-invasive techniques for assessing nerve fiber orientation in the human brain by using diffusion MRI. However, the manual extraction of nerve fiber directions of histological slices is tedious, time consuming, and prone to human error. The presented semi-automated algorithm first creates a binary-segmented mask of the nerve fibers in the histological image, and then extracts an estimate of average directionality of nerve fibers through a Fourier-domain analysis of the masked image. It also generates an uncertainty level for its estimate of average directionality. The average orientations of the semi-automatic method were first compared to a qualitative expert opinion based on visual inspection of nerve fibers. A weighted RMS difference between the expert estimate and the algorithmically determined angle (weighted by expert's confidence in his estimate) was 15.4°, dropping to 9.9° when only cases with an expert confidence level of greater than 50% were included. The algorithmically determined angles were then compared with angles extracted using a manual segmentation technique, yielding an RMS difference of 11.2°. The presented semi-automated method is in good agreement with both qualitative and quantitative manual expert-based approaches for estimating directionality of nerve fibers in white matter from images of stained histological slices of the human brain.