Learning to predict cutting angles from histological human brain sections

Christian Schiffer, Luisa Schuhmacher, Katrin Amunts, Timo Dickscheid

Keywords: cytoarchitecture, histology, deep learning, human brain

TL;DR: We present a method to predict cutting angles of histological brain sections, aiming to exclude obliquely cut tissue regions from subsequent analysis tasks.

Abstract: Studying brain architecture at the cellular level requires histological image analysis of sectioned postmortem samples. We trained a deep neural network to estimate relative angles between the cutting plane and the local 3D brain surface from 2D cortical image patches sampled from microscopic scans of human brain tissue sections. The model allows to automatically identify obliquely cut tissue parts, which often confuse downstream texture classification tasks and typically require specific treatment in image analysis workflows. It has immediate applications for the automated analysis of brain structures, like cytoarchitectonic mapping of the highly convoluted human brain.

Paper Type: validation/application paper

Primary Subject Area: Application: Other

Secondary Subject Area: Segmentation

Paper Status: original work, not submitted yet

Source Code Url: https://jugit.fz-juelich.de/experiments_cschiffer/2020_cutting_angle

Data Set Url: Images of histological brain sections are currently not publicly available due to infrastructure limitations resulting from the large size and number of files. 3D reconstructed data of the BigBrain used for cutting angle estimation can be found at ftp://bigbrain.loris.ca/BigBrainRelease.2015/.

Registration: I acknowledge that publication of this at MIDL and in the proceedings requires at least one of the authors to register and present the work during the conference.

Authorship: I confirm that I am the author of this work and that it has not been submitted to another publication before.