Texture Analysis as a Tool for Quantifying Cellular Architecture in the Human Brain

2116 

Abstract Submission 

Sebastian Bludau¹, Andrea Brandstetter², Olga Kedo², Hartmut Mohlberg³, Katrin Amunts³

¹Forschungszentrum Julich GmbH / INM-1, Juelich, NRW, ²Forschungszentrum Julich GmbH / INM-1, Juelich, Deutschland, ³Institute of Neuroscience and Medicine (INM-1), Research Centre Jülich, Jülich, Germany

Sebastian Bludau  
Forschungszentrum Julich GmbH / INM-1
Juelich, NRW

Andrea Brandstetter  
Forschungszentrum Julich GmbH / INM-1
Juelich, Deutschland

Olga Kedo  
Forschungszentrum Julich GmbH / INM-1
Juelich, Deutschland

Hartmut Mohlberg  
Institute of Neuroscience and Medicine (INM-1), Research Centre Jülich
Jülich, Germany

Katrin Amunts  
Institute of Neuroscience and Medicine (INM-1), Research Centre Jülich
Jülich, Germany

Advancements in optical methods, image analysis and 3D mapping have improved our understanding of the brain's microstructure in the past years. Different methods have been proposed to analyze the architecture of the cerebral cortex, e.g., extracting line profiles to quantify laminar changes in cell density (e.g., [1]), or descriptive approaches for characterizing subcortical nuclei (e.g., mean cell densities, cell sizes). However, there is a lack of methods to reproducibly describe the architecture of the numerous and highly heterogeneous areas in the human brain in a more comprehensive way that considers variations in cellular distributions. Here we introduce texture feature analysis in high-resolution histological images to characterize the brain's architecture. As a proof-of-principle, it has been applied to histological images of the lateral and medial geniculate bodies (CGL and CGM)[2].

We analyzed 550 regions of interest from 10 human postmortem brain (resolution 1µm in-plane). We leveraged the Julich Brain Atlas [3] for anatomical reference. The texture analysis was based on the Gray Level Co-occurrence Matrix (GLCM) method [4] that evaluates the frequency and relationship of pixel intensity pairs neighborhoods across the images. From the GLCM, 21 modified Haralick texture features were extracted, encompassing aspects like contrast, homogeneity, energy, and correlation [4,5] to quantify the cellular architecture of the tissue. A Principal Component Analysis (PCA) reduced these features to four main components, capturing 96.18% of total variance. The distinctiveness between the CGL and CGM was assessed using Independent-samples Kruskal-Wallis tests. Additionally, we examined the intra-structural differences within the CGL's six laminae, demonstrating the method's applicability to more complex neurohistological structures (Fig 1 a,b).

The texture analysis resulted in a separation of the CGL and CGM, showing unique microstructural traits in both areas. A subsequent PCA effectively simplified the texture data, and especially the first four components stood out in how clearly they differentiated from each other. Within the CGL, we observed significant textural differences between the layers, especially between magnocellular (Lamina 1-2) and parvocellular layers (Lamina 3-6). These differences were statistically significant, with the second and third PCA components showing substantial variations across the laminae (Fig 2 a,b).

Texture analysis of high-res histological images quantified the brain's complex micro-architecture and parcellation going beyond basic metrics like cell density. Therefore, it seems to be a promising tool to provide new insights into the brain's architecture, both in cortical and subcortical structures. This approach excels in discerning both notable distinctions, like those between CGL and CGM, and subtler variations within CGL's magno- and parvocellular layers. The interpretation of texture features in terms of traditional histological features is sometimes not straightforward, e.g., due to the different contributions of features to PCA factors, resulting in each factor representing a mix of features. However, such complexity corroborates with microscopic observations, especially in highly heterogeneous subcortical nuclei. The present study provided first evidence that texture analysis enables a reproducible and quantitative characterization of cellular architecture, deepening our grasp on brain organization.

Segmentation and Parcellation ²

Cortical Anatomy and Brain Mapping ¹

Brain Atlases

Cortex

Data analysis

Sub-Cortical

Other - texture analysis