Dilated dense U-net for infant hippocampus subfield segmentation

Hancan Zhu, Feng Shi, Li Wang, Sheng Che Hung, Meng Hsiang Chen, Shuai Wang, Weili Lin, Dinggang Shen

Research output: Contribution to journalArticle

3 Citations (Scopus)

Abstract

Accurate and automatic segmentation of infant hippocampal subfields from magnetic resonance (MR) images is an important step for studying memory related infant neurological diseases. However, existing hippocampal subfield segmentation methods were generally designed based on adult subjects, and would compromise performance when applied to infant subjects due to insufficient tissue contrast and fast changing structural patterns of early hippocampal development. In this paper, we propose a new fully convolutional network (FCN) for infant hippocampal subfield segmentation by embedding the dilated dense network in the U-net, namely DUnet. The embedded dilated dense network can generate multi-scale features while keeping high spatial resolution, which is useful in fusing the low-level features in the contracting path with the high-level features in the expanding path. To further improve the performance, we group every pair of convolutional layers with one residual connection in the DUnet, and obtain the Residual DUnet (ResDUnet). Experimental results show that our proposed DUnet and ResDUnet improve the average Dice coefficient by 2.1 and 2.5% for infant hippocampal subfield segmentation, respectively, when compared with the classic 3D U-net. The results also demonstrate that our methods outperform other state-of-the-art methods.

Original languageEnglish
Article number30
JournalFrontiers in Neuroinformatics
Volume13
DOIs
Publication statusPublished - 2019 Apr 16

Fingerprint

Magnetic resonance
Hippocampus
Tissue
Data storage equipment
Magnetic Resonance Spectroscopy

Keywords

  • Deep learning
  • Dilated dense network
  • Fully convolutional network
  • Hippocampal subfield segmentation
  • Infant hippocampus

ASJC Scopus subject areas

  • Neuroscience (miscellaneous)
  • Biomedical Engineering
  • Computer Science Applications

Cite this

Dilated dense U-net for infant hippocampus subfield segmentation. / Zhu, Hancan; Shi, Feng; Wang, Li; Hung, Sheng Che; Chen, Meng Hsiang; Wang, Shuai; Lin, Weili; Shen, Dinggang.

In: Frontiers in Neuroinformatics, Vol. 13, 30, 16.04.2019.

Research output: Contribution to journalArticle

Zhu, Hancan ; Shi, Feng ; Wang, Li ; Hung, Sheng Che ; Chen, Meng Hsiang ; Wang, Shuai ; Lin, Weili ; Shen, Dinggang. / Dilated dense U-net for infant hippocampus subfield segmentation. In: Frontiers in Neuroinformatics. 2019 ; Vol. 13.
@article{93d09d211ef645a6aefa272cac411125,
title = "Dilated dense U-net for infant hippocampus subfield segmentation",
abstract = "Accurate and automatic segmentation of infant hippocampal subfields from magnetic resonance (MR) images is an important step for studying memory related infant neurological diseases. However, existing hippocampal subfield segmentation methods were generally designed based on adult subjects, and would compromise performance when applied to infant subjects due to insufficient tissue contrast and fast changing structural patterns of early hippocampal development. In this paper, we propose a new fully convolutional network (FCN) for infant hippocampal subfield segmentation by embedding the dilated dense network in the U-net, namely DUnet. The embedded dilated dense network can generate multi-scale features while keeping high spatial resolution, which is useful in fusing the low-level features in the contracting path with the high-level features in the expanding path. To further improve the performance, we group every pair of convolutional layers with one residual connection in the DUnet, and obtain the Residual DUnet (ResDUnet). Experimental results show that our proposed DUnet and ResDUnet improve the average Dice coefficient by 2.1 and 2.5{\%} for infant hippocampal subfield segmentation, respectively, when compared with the classic 3D U-net. The results also demonstrate that our methods outperform other state-of-the-art methods.",
keywords = "Deep learning, Dilated dense network, Fully convolutional network, Hippocampal subfield segmentation, Infant hippocampus",
author = "Hancan Zhu and Feng Shi and Li Wang and Hung, {Sheng Che} and Chen, {Meng Hsiang} and Shuai Wang and Weili Lin and Dinggang Shen",
year = "2019",
month = "4",
day = "16",
doi = "10.3389/fninf.2019.00030",
language = "English",
volume = "13",
journal = "Frontiers in Neuroinformatics",
issn = "1662-5196",
publisher = "Frontiers Research Foundation",

}

TY - JOUR

T1 - Dilated dense U-net for infant hippocampus subfield segmentation

AU - Zhu, Hancan

AU - Shi, Feng

AU - Wang, Li

AU - Hung, Sheng Che

AU - Chen, Meng Hsiang

AU - Wang, Shuai

AU - Lin, Weili

AU - Shen, Dinggang

PY - 2019/4/16

Y1 - 2019/4/16

N2 - Accurate and automatic segmentation of infant hippocampal subfields from magnetic resonance (MR) images is an important step for studying memory related infant neurological diseases. However, existing hippocampal subfield segmentation methods were generally designed based on adult subjects, and would compromise performance when applied to infant subjects due to insufficient tissue contrast and fast changing structural patterns of early hippocampal development. In this paper, we propose a new fully convolutional network (FCN) for infant hippocampal subfield segmentation by embedding the dilated dense network in the U-net, namely DUnet. The embedded dilated dense network can generate multi-scale features while keeping high spatial resolution, which is useful in fusing the low-level features in the contracting path with the high-level features in the expanding path. To further improve the performance, we group every pair of convolutional layers with one residual connection in the DUnet, and obtain the Residual DUnet (ResDUnet). Experimental results show that our proposed DUnet and ResDUnet improve the average Dice coefficient by 2.1 and 2.5% for infant hippocampal subfield segmentation, respectively, when compared with the classic 3D U-net. The results also demonstrate that our methods outperform other state-of-the-art methods.

AB - Accurate and automatic segmentation of infant hippocampal subfields from magnetic resonance (MR) images is an important step for studying memory related infant neurological diseases. However, existing hippocampal subfield segmentation methods were generally designed based on adult subjects, and would compromise performance when applied to infant subjects due to insufficient tissue contrast and fast changing structural patterns of early hippocampal development. In this paper, we propose a new fully convolutional network (FCN) for infant hippocampal subfield segmentation by embedding the dilated dense network in the U-net, namely DUnet. The embedded dilated dense network can generate multi-scale features while keeping high spatial resolution, which is useful in fusing the low-level features in the contracting path with the high-level features in the expanding path. To further improve the performance, we group every pair of convolutional layers with one residual connection in the DUnet, and obtain the Residual DUnet (ResDUnet). Experimental results show that our proposed DUnet and ResDUnet improve the average Dice coefficient by 2.1 and 2.5% for infant hippocampal subfield segmentation, respectively, when compared with the classic 3D U-net. The results also demonstrate that our methods outperform other state-of-the-art methods.

KW - Deep learning

KW - Dilated dense network

KW - Fully convolutional network

KW - Hippocampal subfield segmentation

KW - Infant hippocampus

UR - http://www.scopus.com/inward/record.url?scp=85067412559&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=85067412559&partnerID=8YFLogxK

U2 - 10.3389/fninf.2019.00030

DO - 10.3389/fninf.2019.00030

M3 - Article

AN - SCOPUS:85067412559

VL - 13

JO - Frontiers in Neuroinformatics

JF - Frontiers in Neuroinformatics

SN - 1662-5196

M1 - 30

ER -