Residual encoder and convolutional decoder neural network for glioma segmentation

Kamlesh Pawar, Zhaolin Chen, N. Jon Shah, Gary Egan

Research output: Chapter in Book/Report/Conference proceedingConference PaperResearchpeer-review

Abstract

A deep learning approach to glioma segmentation is presented. An encoder and decoder pair deep learning network is designed which takes T1, T2, T1-CE (contrast enhanced) and T2-Flair (fluid attenuation inversion recovery) images as input and outputs the segmented labels. The encoder is a 49 layer deep residual learning architecture that encodes the 240 × 240 × 4 input images into 8 × 8 × 2048 feature maps. The decoder network takes these feature maps and extract the segmented labels. The decoder network is fully convolutional network consisting of convolutional and upsampling layers. Additionally, the input images are downsampled using bilinear interpolation and are inserted into the decoder network through concatenation. This concatenation step provides spatial information of the tumor to the decoder, which was lost due to pooling/downlsampling during encoding. The network is trained on the BRATS-17 training dataset and validated on the validation dataset. The dice score, sensitivity and specificity of the segmented whole tumor, core tumor and enhancing tumor is computed on validation dataset. The mean dice score for whole tumor, core tumor and enhancing tumor for validation dataset were 0.824, 0.627 and 0.575, respectively.
Original languageEnglish
Title of host publicationBrainlesion
Subtitle of host publicationGlioma, Multiple Sclerosis, Stroke and Traumatic Brain Injuries - 3rd International Workshop, BrainLes 2017, Held in Conjunction with MICCAI 2017, Revised Selected Papers
Place of PublicationSwitzerland
PublisherSpringer
Pages263-273
Number of pages11
ISBN (Electronic)9783319752389
ISBN (Print)9783319752372
DOIs
Publication statusPublished - 1 Jan 2018
EventMedical Image Computing and Computer-Assisted Intervention 2017 - Quebec City Convention Centre, Quebec, Canada
Duration: 10 Sep 201714 Sep 2017
Conference number: 20th
http://www.miccai2017.org/

Publication series

NameLecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Volume10670 LNCS
ISSN (Print)0302-9743
ISSN (Electronic)1611-3349

Conference

ConferenceMedical Image Computing and Computer-Assisted Intervention 2017
Abbreviated titleMICCAI 2017
CountryCanada
CityQuebec
Period10/09/1714/09/17
Internet address

Keywords

  • CNN
  • Computer vision
  • Deep learning
  • Image segmentation

Cite this

Pawar, K., Chen, Z., Shah, N. J., & Egan, G. (2018). Residual encoder and convolutional decoder neural network for glioma segmentation. In Brainlesion: Glioma, Multiple Sclerosis, Stroke and Traumatic Brain Injuries - 3rd International Workshop, BrainLes 2017, Held in Conjunction with MICCAI 2017, Revised Selected Papers (pp. 263-273). (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); Vol. 10670 LNCS). Switzerland: Springer. https://doi.org/10.1007/978-3-319-75238-9_23
Pawar, Kamlesh ; Chen, Zhaolin ; Shah, N. Jon ; Egan, Gary. / Residual encoder and convolutional decoder neural network for glioma segmentation. Brainlesion: Glioma, Multiple Sclerosis, Stroke and Traumatic Brain Injuries - 3rd International Workshop, BrainLes 2017, Held in Conjunction with MICCAI 2017, Revised Selected Papers. Switzerland : Springer, 2018. pp. 263-273 (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)).
@inproceedings{f76e7a91150d4577a67ad3d77dc0f70d,
title = "Residual encoder and convolutional decoder neural network for glioma segmentation",
abstract = "A deep learning approach to glioma segmentation is presented. An encoder and decoder pair deep learning network is designed which takes T1, T2, T1-CE (contrast enhanced) and T2-Flair (fluid attenuation inversion recovery) images as input and outputs the segmented labels. The encoder is a 49 layer deep residual learning architecture that encodes the 240 × 240 × 4 input images into 8 × 8 × 2048 feature maps. The decoder network takes these feature maps and extract the segmented labels. The decoder network is fully convolutional network consisting of convolutional and upsampling layers. Additionally, the input images are downsampled using bilinear interpolation and are inserted into the decoder network through concatenation. This concatenation step provides spatial information of the tumor to the decoder, which was lost due to pooling/downlsampling during encoding. The network is trained on the BRATS-17 training dataset and validated on the validation dataset. The dice score, sensitivity and specificity of the segmented whole tumor, core tumor and enhancing tumor is computed on validation dataset. The mean dice score for whole tumor, core tumor and enhancing tumor for validation dataset were 0.824, 0.627 and 0.575, respectively.",
keywords = "CNN, Computer vision, Deep learning, Image segmentation",
author = "Kamlesh Pawar and Zhaolin Chen and Shah, {N. Jon} and Gary Egan",
year = "2018",
month = "1",
day = "1",
doi = "10.1007/978-3-319-75238-9_23",
language = "English",
isbn = "9783319752372",
series = "Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)",
publisher = "Springer",
pages = "263--273",
booktitle = "Brainlesion",

}

Pawar, K, Chen, Z, Shah, NJ & Egan, G 2018, Residual encoder and convolutional decoder neural network for glioma segmentation. in Brainlesion: Glioma, Multiple Sclerosis, Stroke and Traumatic Brain Injuries - 3rd International Workshop, BrainLes 2017, Held in Conjunction with MICCAI 2017, Revised Selected Papers. Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics), vol. 10670 LNCS, Springer, Switzerland, pp. 263-273, Medical Image Computing and Computer-Assisted Intervention 2017, Quebec, Canada, 10/09/17. https://doi.org/10.1007/978-3-319-75238-9_23

Residual encoder and convolutional decoder neural network for glioma segmentation. / Pawar, Kamlesh; Chen, Zhaolin; Shah, N. Jon; Egan, Gary.

Brainlesion: Glioma, Multiple Sclerosis, Stroke and Traumatic Brain Injuries - 3rd International Workshop, BrainLes 2017, Held in Conjunction with MICCAI 2017, Revised Selected Papers. Switzerland : Springer, 2018. p. 263-273 (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); Vol. 10670 LNCS).

Research output: Chapter in Book/Report/Conference proceedingConference PaperResearchpeer-review

TY - GEN

T1 - Residual encoder and convolutional decoder neural network for glioma segmentation

AU - Pawar, Kamlesh

AU - Chen, Zhaolin

AU - Shah, N. Jon

AU - Egan, Gary

PY - 2018/1/1

Y1 - 2018/1/1

N2 - A deep learning approach to glioma segmentation is presented. An encoder and decoder pair deep learning network is designed which takes T1, T2, T1-CE (contrast enhanced) and T2-Flair (fluid attenuation inversion recovery) images as input and outputs the segmented labels. The encoder is a 49 layer deep residual learning architecture that encodes the 240 × 240 × 4 input images into 8 × 8 × 2048 feature maps. The decoder network takes these feature maps and extract the segmented labels. The decoder network is fully convolutional network consisting of convolutional and upsampling layers. Additionally, the input images are downsampled using bilinear interpolation and are inserted into the decoder network through concatenation. This concatenation step provides spatial information of the tumor to the decoder, which was lost due to pooling/downlsampling during encoding. The network is trained on the BRATS-17 training dataset and validated on the validation dataset. The dice score, sensitivity and specificity of the segmented whole tumor, core tumor and enhancing tumor is computed on validation dataset. The mean dice score for whole tumor, core tumor and enhancing tumor for validation dataset were 0.824, 0.627 and 0.575, respectively.

AB - A deep learning approach to glioma segmentation is presented. An encoder and decoder pair deep learning network is designed which takes T1, T2, T1-CE (contrast enhanced) and T2-Flair (fluid attenuation inversion recovery) images as input and outputs the segmented labels. The encoder is a 49 layer deep residual learning architecture that encodes the 240 × 240 × 4 input images into 8 × 8 × 2048 feature maps. The decoder network takes these feature maps and extract the segmented labels. The decoder network is fully convolutional network consisting of convolutional and upsampling layers. Additionally, the input images are downsampled using bilinear interpolation and are inserted into the decoder network through concatenation. This concatenation step provides spatial information of the tumor to the decoder, which was lost due to pooling/downlsampling during encoding. The network is trained on the BRATS-17 training dataset and validated on the validation dataset. The dice score, sensitivity and specificity of the segmented whole tumor, core tumor and enhancing tumor is computed on validation dataset. The mean dice score for whole tumor, core tumor and enhancing tumor for validation dataset were 0.824, 0.627 and 0.575, respectively.

KW - CNN

KW - Computer vision

KW - Deep learning

KW - Image segmentation

UR - http://www.scopus.com/inward/record.url?scp=85042544528&partnerID=8YFLogxK

U2 - 10.1007/978-3-319-75238-9_23

DO - 10.1007/978-3-319-75238-9_23

M3 - Conference Paper

SN - 9783319752372

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

SP - 263

EP - 273

BT - Brainlesion

PB - Springer

CY - Switzerland

ER -

Pawar K, Chen Z, Shah NJ, Egan G. Residual encoder and convolutional decoder neural network for glioma segmentation. In Brainlesion: Glioma, Multiple Sclerosis, Stroke and Traumatic Brain Injuries - 3rd International Workshop, BrainLes 2017, Held in Conjunction with MICCAI 2017, Revised Selected Papers. Switzerland: Springer. 2018. p. 263-273. (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)). https://doi.org/10.1007/978-3-319-75238-9_23