Semi-supervised deep learning of brain tissue segmentation

Ryo Ito, Ken Nakae, Junichi Hata, Hideyuki Okano, Shin Ishii

Research output: Contribution to journalArticle

Abstract

Brain image segmentation is of great importance not only for clinical use but also for neuroscience research. Recent developments in deep neural networks (DNNs) have led to the application of DNNs to brain image segmentation, which required extensive human annotations of whole brain images. Annotating three-dimensional brain images requires laborious efforts by expert anatomists because of the differences among images in terms of their dimensionality, noise, contrast, or ambiguous boundaries that even prevent these experts from necessarily attaining consistency. This paper proposes a semi-supervised learning framework to train a DNN based on a relatively small number of annotated (labeled) images, named atlases, but also a relatively large number of unlabeled images by leveraging image registration to attach pseudo-labels to images that were originally unlabeled. We applied our proposed method to two different datasets: open human brain images and our original marmoset brain images. When provided with the same number of atlases for training, we found our method achieved superior and more stable segmentation results than those by existing registration-based and DNN-based methods.

Original languageEnglish
Pages (from-to)25-34
Number of pages10
JournalNeural Networks
Volume116
DOIs
Publication statusPublished - 2019 Aug 1

Fingerprint

Brain
Tissue
Atlases
Image segmentation
Anatomists
Callithrix
Three-Dimensional Imaging
Image registration
Supervised learning
Neurosciences
Noise
Deep learning
Supervised Machine Learning
Labels
Deep neural networks
Research

Keywords

  • Brain tissue segmentation
  • Deep neural network
  • Image registration
  • Semi-supervised learning

ASJC Scopus subject areas

  • Cognitive Neuroscience
  • Artificial Intelligence

Cite this

Semi-supervised deep learning of brain tissue segmentation. / Ito, Ryo; Nakae, Ken; Hata, Junichi; Okano, Hideyuki; Ishii, Shin.

In: Neural Networks, Vol. 116, 01.08.2019, p. 25-34.

Research output: Contribution to journalArticle

Ito, Ryo ; Nakae, Ken ; Hata, Junichi ; Okano, Hideyuki ; Ishii, Shin. / Semi-supervised deep learning of brain tissue segmentation. In: Neural Networks. 2019 ; Vol. 116. pp. 25-34.
@article{665f766acdba4352b03eab6ecdcd3bf1,
title = "Semi-supervised deep learning of brain tissue segmentation",
abstract = "Brain image segmentation is of great importance not only for clinical use but also for neuroscience research. Recent developments in deep neural networks (DNNs) have led to the application of DNNs to brain image segmentation, which required extensive human annotations of whole brain images. Annotating three-dimensional brain images requires laborious efforts by expert anatomists because of the differences among images in terms of their dimensionality, noise, contrast, or ambiguous boundaries that even prevent these experts from necessarily attaining consistency. This paper proposes a semi-supervised learning framework to train a DNN based on a relatively small number of annotated (labeled) images, named atlases, but also a relatively large number of unlabeled images by leveraging image registration to attach pseudo-labels to images that were originally unlabeled. We applied our proposed method to two different datasets: open human brain images and our original marmoset brain images. When provided with the same number of atlases for training, we found our method achieved superior and more stable segmentation results than those by existing registration-based and DNN-based methods.",
keywords = "Brain tissue segmentation, Deep neural network, Image registration, Semi-supervised learning",
author = "Ryo Ito and Ken Nakae and Junichi Hata and Hideyuki Okano and Shin Ishii",
year = "2019",
month = "8",
day = "1",
doi = "10.1016/j.neunet.2019.03.014",
language = "English",
volume = "116",
pages = "25--34",
journal = "Neural Networks",
issn = "0893-6080",
publisher = "Elsevier Limited",

}

TY - JOUR

T1 - Semi-supervised deep learning of brain tissue segmentation

AU - Ito, Ryo

AU - Nakae, Ken

AU - Hata, Junichi

AU - Okano, Hideyuki

AU - Ishii, Shin

PY - 2019/8/1

Y1 - 2019/8/1

N2 - Brain image segmentation is of great importance not only for clinical use but also for neuroscience research. Recent developments in deep neural networks (DNNs) have led to the application of DNNs to brain image segmentation, which required extensive human annotations of whole brain images. Annotating three-dimensional brain images requires laborious efforts by expert anatomists because of the differences among images in terms of their dimensionality, noise, contrast, or ambiguous boundaries that even prevent these experts from necessarily attaining consistency. This paper proposes a semi-supervised learning framework to train a DNN based on a relatively small number of annotated (labeled) images, named atlases, but also a relatively large number of unlabeled images by leveraging image registration to attach pseudo-labels to images that were originally unlabeled. We applied our proposed method to two different datasets: open human brain images and our original marmoset brain images. When provided with the same number of atlases for training, we found our method achieved superior and more stable segmentation results than those by existing registration-based and DNN-based methods.

AB - Brain image segmentation is of great importance not only for clinical use but also for neuroscience research. Recent developments in deep neural networks (DNNs) have led to the application of DNNs to brain image segmentation, which required extensive human annotations of whole brain images. Annotating three-dimensional brain images requires laborious efforts by expert anatomists because of the differences among images in terms of their dimensionality, noise, contrast, or ambiguous boundaries that even prevent these experts from necessarily attaining consistency. This paper proposes a semi-supervised learning framework to train a DNN based on a relatively small number of annotated (labeled) images, named atlases, but also a relatively large number of unlabeled images by leveraging image registration to attach pseudo-labels to images that were originally unlabeled. We applied our proposed method to two different datasets: open human brain images and our original marmoset brain images. When provided with the same number of atlases for training, we found our method achieved superior and more stable segmentation results than those by existing registration-based and DNN-based methods.

KW - Brain tissue segmentation

KW - Deep neural network

KW - Image registration

KW - Semi-supervised learning

UR - http://www.scopus.com/inward/record.url?scp=85064248652&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=85064248652&partnerID=8YFLogxK

U2 - 10.1016/j.neunet.2019.03.014

DO - 10.1016/j.neunet.2019.03.014

M3 - Article

VL - 116

SP - 25

EP - 34

JO - Neural Networks

JF - Neural Networks

SN - 0893-6080

ER -