Independent Component Analysis for Noise and Artifact Removal in Three-Dimensional Polarized Light Imaging

Abstract

In recent years, Independent Component Analysis (ICA) has successfully been applied to remove noise and artifacts in images obtained from Three-dimensional Polarized Light Imaging (3D-PLI) at the mesoscale (i.e., 64 $$\upmu $$ μ m). Here, we present an automatic denoising procedure for gray matter regions that allows to apply the ICA also to microscopic images, with reasonable computational effort. Apart from an automatic segmentation of gray matter regions, we applied the denoising procedure to several 3D-PLI images from a rat and a vervet monkey brain section.

Automated computation of nerve fibre inclinations from 3D polarised light imaging measurements of brain tissue

“…In addition to the fibre inclinations, we provided an HM-probability map, indicating the probability that a region is highly myelinated. Benning et al 21 proposed a binary separation of regions with low and high myelination. We improved the existing algorithm (taking different features/plateaus of the histograms into account and excluding background tissue when determining the threshold parameters).…”

“…We followed the classification introduced in Benning et al 21 to separate regions with low and high myelination (LM- and HM-regions). Note that the authors used the terms “grey matter” and “white matter”, while we use the terms LM- and HM-regions to avoid confusion with the anatomical definitions.…”

“…As in Benning et al 21 , the threshold parameters and were computed as the points of maximum curvature behind and before the biggest peak in the retardation and transmittance histogram, respectively. In contrast to Benning et al 21 who used 128 bins, we decided to use histograms with 256 bins in order to increase accuracy. However, histograms with 256 bins might show several small peaks or a plateau (cf.…”

“…2 b on the very left), which makes the determination of the threshold parameters difficult. To ensure that no outliers were chosen, we first considered a histogram with 64 bins and computed the threshold parameters (points of maximum curvature) in a range corresponding to the full-width at half-maximum (FWHM) of the peaks, using larger ranges as Benning et al 21 to account for thinner peaks: 20 FWHM behind the retardation peak and 10 FWHM before the transmittance peak. Then, we considered histograms with 128 bins and recomputed the threshold parameters in a range of bins close to the previous results [ ].…”

“…In regions with a low degree of myelination, an unweighted model is used. The separation of regions with low and high degrees of myelination is based on an algorithm proposed by Benning et al 21 . However, instead of using a binary classification, we introduce the important concept of transition zones, allowing to consider subtle changes in the tissue composition and to compute more accurate fibre inclinations at boundaries.…”