Cytoarchitectonic parcellations of the human brain serve as anatomical references in multimodal atlas frameworks. They are based on analysis of cell-body stained histological sections and the identification of borders between brain areas. The de-facto standard involves a semi-automatic, reproducible border detection, but does not scale with high-throughput imaging in large series of sections at microscopical resolution. Automatic parcellation, however, is extremely challenging due to high variation in the data, and the need for a large field of view at microscopic resolution. The performance of a recently proposed Convolutional Neural Network model that addresses this problem especially suffers from the naturally limited amount of expert annotations for training. To circumvent this limitation, we propose to pre-train neural networks on a self-supervised auxiliary task, predicting the 3D distance between two patches sampled from the same brain. Compared to a random initialization, fine-tuning from these networks results in significantly better segmentations. We show that the self-supervised model has implicitly learned to distinguish several cortical brain areas -- a strong indicator that the proposed auxiliary task is appropriate for cytoarchitectonic mapping.
Description
Improving Cytoarchitectonic Segmentation of Human Brain Areas with Self-supervised Siamese Networks | SpringerLink
%0 Conference Paper
%1 10.1007/978-3-030-00931-1_76
%A Spitzer, Hannah
%A Kiwitz, Kai
%A Amunts, Katrin
%A Harmeling, Stefan
%A Dickscheid, Timo
%B Medical Image Computing and Computer Assisted Intervention -- MICCAI 2018
%C Cham
%D 2018
%E Frangi, Alejandro F.
%E Schnabel, Julia A.
%E Davatzikos, Christos
%E Alberola-López, Carlos
%E Fichtinger, Gabor
%I Springer International Publishing
%K myown
%P 663--671
%T Improving Cytoarchitectonic Segmentation of Human Brain Areas with Self-supervised Siamese Networks
%X Cytoarchitectonic parcellations of the human brain serve as anatomical references in multimodal atlas frameworks. They are based on analysis of cell-body stained histological sections and the identification of borders between brain areas. The de-facto standard involves a semi-automatic, reproducible border detection, but does not scale with high-throughput imaging in large series of sections at microscopical resolution. Automatic parcellation, however, is extremely challenging due to high variation in the data, and the need for a large field of view at microscopic resolution. The performance of a recently proposed Convolutional Neural Network model that addresses this problem especially suffers from the naturally limited amount of expert annotations for training. To circumvent this limitation, we propose to pre-train neural networks on a self-supervised auxiliary task, predicting the 3D distance between two patches sampled from the same brain. Compared to a random initialization, fine-tuning from these networks results in significantly better segmentations. We show that the self-supervised model has implicitly learned to distinguish several cortical brain areas -- a strong indicator that the proposed auxiliary task is appropriate for cytoarchitectonic mapping.
%@ 978-3-030-00931-1
@inproceedings{10.1007/978-3-030-00931-1_76,
abstract = {Cytoarchitectonic parcellations of the human brain serve as anatomical references in multimodal atlas frameworks. They are based on analysis of cell-body stained histological sections and the identification of borders between brain areas. The de-facto standard involves a semi-automatic, reproducible border detection, but does not scale with high-throughput imaging in large series of sections at microscopical resolution. Automatic parcellation, however, is extremely challenging due to high variation in the data, and the need for a large field of view at microscopic resolution. The performance of a recently proposed Convolutional Neural Network model that addresses this problem especially suffers from the naturally limited amount of expert annotations for training. To circumvent this limitation, we propose to pre-train neural networks on a self-supervised auxiliary task, predicting the 3D distance between two patches sampled from the same brain. Compared to a random initialization, fine-tuning from these networks results in significantly better segmentations. We show that the self-supervised model has implicitly learned to distinguish several cortical brain areas -- a strong indicator that the proposed auxiliary task is appropriate for cytoarchitectonic mapping.},
added-at = {2019-04-09T11:38:55.000+0200},
address = {Cham},
author = {Spitzer, Hannah and Kiwitz, Kai and Amunts, Katrin and Harmeling, Stefan and Dickscheid, Timo},
biburl = {https://www.bibsonomy.org/bibtex/2be810f7e010a2a6dca3dbe9e4d78a1bf/dickscheid},
booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2018},
description = {Improving Cytoarchitectonic Segmentation of Human Brain Areas with Self-supervised Siamese Networks | SpringerLink},
editor = {Frangi, Alejandro F. and Schnabel, Julia A. and Davatzikos, Christos and Alberola-L{\'o}pez, Carlos and Fichtinger, Gabor},
interhash = {31a8619e74245151a58ed8dcd1175bb5},
intrahash = {be810f7e010a2a6dca3dbe9e4d78a1bf},
isbn = {978-3-030-00931-1},
keywords = {myown},
pages = {663--671},
publisher = {Springer International Publishing},
timestamp = {2019-04-09T15:39:51.000+0200},
title = {Improving Cytoarchitectonic Segmentation of Human Brain Areas with Self-supervised Siamese Networks},
year = 2018
}