Toward Automated Region Detection & Parcellation of Rat Brain Tissue Images
Abstract
People who analyze images of biological tissue often rely on segmentation of structures as a preliminary step. In particular, laboratories studying the rat brain manually delineate brain regions to position scientific findings on a brain atlas to propose hypotheses about the rat brain, and ultimately, the human brain. Our work intersects with the preliminary step of delineating regions in images of brain tissue via computational methods. We investigate pixel-wise classification or segmentation of brain regions using ten histological images of brain tissue sections stained for Nissl substance, and two deep learning models: U-Net and Tile2Vec. Our goal is to assess the viability of segmenting brain structures from images alone with both supervised and unsupervised approaches. Further, we determine how image resolution and additional domain knowledge affects segmentation. Experimenting with different resolutions shows the supervised model performs best when the data has enough resolution to distinguish cytoarchitectural patterns. At the same time additional domain knowledge, in the form of atlas-guided parcellations, improves segmentation for some cases while misclassification occurs in other cases. We argue misclassification is partly due to limited availability of data, rendering the supervised model incapable of performing well on data it has never seen. To this end, we employed an unsupervised approach where the goal is to generate lower dimensional representations of image tiles taken from the histological images. This approach is data efficient and enables segmentation of different structures with traditional computer vision techniques. Overall, our work shows segmenting structures is possible with histological images of brain tissue sections stained for Nissl substance. Our efforts contribute to a long history of characterizing the brain. We continue working with intentions of, one day, streamlining reconstruction of rat brain maps, and other animal models, using histological data.
Subject Area
Applied Mathematics|Neurosciences|Bioinformatics
Recommended Citation
Arnal, Alexandro, "Toward Automated Region Detection & Parcellation of Rat Brain Tissue Images" (2020). ETD Collection for University of Texas, El Paso. AAI27999900.
https://scholarworks.utep.edu/dissertations/AAI27999900