Representation learning: A unified deep learning framework for automatic prostate MR segmentation

Shu Liao, Yaozong Gao, Aytekin Oto, Dinggang Shen

Research output: Chapter in Book/Report/Conference proceedingConference contribution

Abstract

Image representation plays an important role in medical image analysis. The key to the success of different medical image analysis algorithms is heavily dependent on how we represent the input data, namely features used to characterize the input image. In the literature, feature engineering remains as an active research topic, and many novel hand-crafted features are designed such as Haar wavelet, histogram of oriented gradient, and local binary patterns. However, such features are not designed with the guidance of the underlying dataset at hand. To this end, we argue that the most effective features should be designed in a learning based manner, namely representation learning, which can be adapted to different patient datasets at hand. In this paper, we introduce a deep learning framework to achieve this goal. Specifically, a stacked independent subspace analysis (ISA) network is adopted to learn the most effective features in a hierarchical and unsupervised manner. The learnt features are adapted to the dataset at hand and encode high level semantic anatomical information. The proposed method is evaluated on the application of automatic prostate MR segmentation. Experimental results show that significant segmentation accuracy improvement can be achieved by the proposed deep learning method compared to other state-of-the-art segmentation approaches.

Original languageEnglish
Title of host publicationLecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Pages254-261
Number of pages8
Volume8150 LNCS
EditionPART 2
DOIs
Publication statusPublished - 2013 Oct 24
Externally publishedYes
Event16th International Conference on Medical Image Computing and Computer Assisted Intervention, MICCAI 2013 - Nagoya, Japan
Duration: 2013 Sep 222013 Sep 26

Publication series

NameLecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
NumberPART 2
Volume8150 LNCS
ISSN (Print)03029743
ISSN (Electronic)16113349

Other

Other16th International Conference on Medical Image Computing and Computer Assisted Intervention, MICCAI 2013
CountryJapan
CityNagoya
Period13/9/2213/9/26

Fingerprint

Image analysis
Segmentation
Medical Image Analysis
Electric network analysis
Semantics
Haar Wavelet
Image Representation
Network Analysis
Histogram
Guidance
Subspace
Binary
Gradient
Engineering
Framework
Learning
Deep learning
Dependent
Experimental Results

ASJC Scopus subject areas

  • Computer Science(all)
  • Theoretical Computer Science

Cite this

Liao, S., Gao, Y., Oto, A., & Shen, D. (2013). Representation learning: A unified deep learning framework for automatic prostate MR segmentation. In Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics) (PART 2 ed., Vol. 8150 LNCS, pp. 254-261). (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); Vol. 8150 LNCS, No. PART 2). https://doi.org/10.1007/978-3-642-40763-5_32

Representation learning : A unified deep learning framework for automatic prostate MR segmentation. / Liao, Shu; Gao, Yaozong; Oto, Aytekin; Shen, Dinggang.

Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics). Vol. 8150 LNCS PART 2. ed. 2013. p. 254-261 (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); Vol. 8150 LNCS, No. PART 2).

Research output: Chapter in Book/Report/Conference proceedingConference contribution

Liao, S, Gao, Y, Oto, A & Shen, D 2013, Representation learning: A unified deep learning framework for automatic prostate MR segmentation. in Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics). PART 2 edn, vol. 8150 LNCS, Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics), no. PART 2, vol. 8150 LNCS, pp. 254-261, 16th International Conference on Medical Image Computing and Computer Assisted Intervention, MICCAI 2013, Nagoya, Japan, 13/9/22. https://doi.org/10.1007/978-3-642-40763-5_32
Liao S, Gao Y, Oto A, Shen D. Representation learning: A unified deep learning framework for automatic prostate MR segmentation. In Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics). PART 2 ed. Vol. 8150 LNCS. 2013. p. 254-261. (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); PART 2). https://doi.org/10.1007/978-3-642-40763-5_32
Liao, Shu ; Gao, Yaozong ; Oto, Aytekin ; Shen, Dinggang. / Representation learning : A unified deep learning framework for automatic prostate MR segmentation. Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics). Vol. 8150 LNCS PART 2. ed. 2013. pp. 254-261 (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); PART 2).
@inproceedings{ed16ee4f7ba3444186ef3a9e6c9f9883,
title = "Representation learning: A unified deep learning framework for automatic prostate MR segmentation",
abstract = "Image representation plays an important role in medical image analysis. The key to the success of different medical image analysis algorithms is heavily dependent on how we represent the input data, namely features used to characterize the input image. In the literature, feature engineering remains as an active research topic, and many novel hand-crafted features are designed such as Haar wavelet, histogram of oriented gradient, and local binary patterns. However, such features are not designed with the guidance of the underlying dataset at hand. To this end, we argue that the most effective features should be designed in a learning based manner, namely representation learning, which can be adapted to different patient datasets at hand. In this paper, we introduce a deep learning framework to achieve this goal. Specifically, a stacked independent subspace analysis (ISA) network is adopted to learn the most effective features in a hierarchical and unsupervised manner. The learnt features are adapted to the dataset at hand and encode high level semantic anatomical information. The proposed method is evaluated on the application of automatic prostate MR segmentation. Experimental results show that significant segmentation accuracy improvement can be achieved by the proposed deep learning method compared to other state-of-the-art segmentation approaches.",
author = "Shu Liao and Yaozong Gao and Aytekin Oto and Dinggang Shen",
year = "2013",
month = "10",
day = "24",
doi = "10.1007/978-3-642-40763-5_32",
language = "English",
isbn = "9783642407628",
volume = "8150 LNCS",
series = "Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)",
number = "PART 2",
pages = "254--261",
booktitle = "Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)",
edition = "PART 2",

}

TY - GEN

T1 - Representation learning

T2 - A unified deep learning framework for automatic prostate MR segmentation

AU - Liao, Shu

AU - Gao, Yaozong

AU - Oto, Aytekin

AU - Shen, Dinggang

PY - 2013/10/24

Y1 - 2013/10/24

N2 - Image representation plays an important role in medical image analysis. The key to the success of different medical image analysis algorithms is heavily dependent on how we represent the input data, namely features used to characterize the input image. In the literature, feature engineering remains as an active research topic, and many novel hand-crafted features are designed such as Haar wavelet, histogram of oriented gradient, and local binary patterns. However, such features are not designed with the guidance of the underlying dataset at hand. To this end, we argue that the most effective features should be designed in a learning based manner, namely representation learning, which can be adapted to different patient datasets at hand. In this paper, we introduce a deep learning framework to achieve this goal. Specifically, a stacked independent subspace analysis (ISA) network is adopted to learn the most effective features in a hierarchical and unsupervised manner. The learnt features are adapted to the dataset at hand and encode high level semantic anatomical information. The proposed method is evaluated on the application of automatic prostate MR segmentation. Experimental results show that significant segmentation accuracy improvement can be achieved by the proposed deep learning method compared to other state-of-the-art segmentation approaches.

AB - Image representation plays an important role in medical image analysis. The key to the success of different medical image analysis algorithms is heavily dependent on how we represent the input data, namely features used to characterize the input image. In the literature, feature engineering remains as an active research topic, and many novel hand-crafted features are designed such as Haar wavelet, histogram of oriented gradient, and local binary patterns. However, such features are not designed with the guidance of the underlying dataset at hand. To this end, we argue that the most effective features should be designed in a learning based manner, namely representation learning, which can be adapted to different patient datasets at hand. In this paper, we introduce a deep learning framework to achieve this goal. Specifically, a stacked independent subspace analysis (ISA) network is adopted to learn the most effective features in a hierarchical and unsupervised manner. The learnt features are adapted to the dataset at hand and encode high level semantic anatomical information. The proposed method is evaluated on the application of automatic prostate MR segmentation. Experimental results show that significant segmentation accuracy improvement can be achieved by the proposed deep learning method compared to other state-of-the-art segmentation approaches.

UR - http://www.scopus.com/inward/record.url?scp=84885909194&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=84885909194&partnerID=8YFLogxK

U2 - 10.1007/978-3-642-40763-5_32

DO - 10.1007/978-3-642-40763-5_32

M3 - Conference contribution

SN - 9783642407628

VL - 8150 LNCS

T3 - Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)

SP - 254

EP - 261

BT - Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)

ER -