Efficient multi-scale 3D CNN with fully connected CRF for accurate brain lesion segmentation
File(s)
Published version
Author(s)
Type
Journal Article
Abstract
We propose a dual pathway, 11-layers deep, three-dimensional Convolutional
Neural Network for the challenging task of brain lesion segmentation. The
devised architecture is the result of an in-depth analysis of the limitations of
current networks proposed for similar applications. To overcome the computational
burden of processing 3D medical scans, we have devised an efficient
and effective dense training scheme which joins the processing of adjacent
image patches into one pass through the network while automatically adapting
to the inherent class imbalance present in the data. Further, we analyze
the development of deeper, thus more discriminative 3D CNNs. In order to
incorporate both local and larger contextual information, we employ a dual
pathway architecture that processes the input images at multiple scales simultaneously.
For post-processing of the network’s soft segmentation, we use a
3D fully connected Conditional Random Field which effectively removes false
positives. Our pipeline is extensively evaluated on three challenging tasks of
lesion segmentation in multi-channel MRI patient data with traumatic brain
injuries, brain tumors, and ischemic stroke. We improve on the state-of-theart
for all three applications, with top ranking performance on the public
benchmarks BRATS 2015 and ISLES 2015. Our method is computationally
efficient, which allows its adoption in a variety of research and clinical
settings. The source code of our implementation is made publicly available
Neural Network for the challenging task of brain lesion segmentation. The
devised architecture is the result of an in-depth analysis of the limitations of
current networks proposed for similar applications. To overcome the computational
burden of processing 3D medical scans, we have devised an efficient
and effective dense training scheme which joins the processing of adjacent
image patches into one pass through the network while automatically adapting
to the inherent class imbalance present in the data. Further, we analyze
the development of deeper, thus more discriminative 3D CNNs. In order to
incorporate both local and larger contextual information, we employ a dual
pathway architecture that processes the input images at multiple scales simultaneously.
For post-processing of the network’s soft segmentation, we use a
3D fully connected Conditional Random Field which effectively removes false
positives. Our pipeline is extensively evaluated on three challenging tasks of
lesion segmentation in multi-channel MRI patient data with traumatic brain
injuries, brain tumors, and ischemic stroke. We improve on the state-of-theart
for all three applications, with top ranking performance on the public
benchmarks BRATS 2015 and ISLES 2015. Our method is computationally
efficient, which allows its adoption in a variety of research and clinical
settings. The source code of our implementation is made publicly available
Date Issued
2017-02-01
Date Acceptance
2016-10-12
Citation
Medical Image Analysis, 2017, 36 (1), pp.61-78
ISSN
1361-8423
Publisher
Elsevier
Start Page
61
End Page
78
Journal / Book Title
Medical Image Analysis
Volume
36
Issue
1
Replaces
10044/1/41612
Copyright Statement
© 2016 The Authors. Published by Elsevier B.V. This is an open access article under the CC BY license (
http://creativecommons.org/licenses/by/4.0/
).
http://creativecommons.org/licenses/by/4.0/
).
License URL
Sponsor
Commission of the European Communities
Engineering & Physical Science Research Council (EPSRC)
Grant Number
HEALTH-F2-2013-602150
EP/N023668/1
Subjects
Science & Technology
Technology
Life Sciences & Biomedicine
Computer Science, Artificial Intelligence
Computer Science, Interdisciplinary Applications
Engineering, Biomedical
Radiology, Nuclear Medicine & Medical Imaging
Computer Science
Engineering
3D convolutional neural network
Fully connected CRF
Segmentation Brain lesions
Deep learning
HIGH-GRADE GLIOMAS
MULTIPLE-SCLEROSIS
DEFORMABLE REGISTRATION
TUMOR SEGMENTATION
INJURY
MR
NETWORKS
IMAGES
3D convolutional neural network
Brain lesions
Deep learning
Fully connected CRF
Segmentation
Brain
Brain Injuries, Traumatic
Brain Ischemia
Brain Neoplasms
Humans
Neural Networks, Computer
Reproducibility of Results
Sensitivity and Specificity
Brain
Humans
Brain Neoplasms
Brain Ischemia
Sensitivity and Specificity
Reproducibility of Results
Brain Injuries, Traumatic
Neural Networks, Computer
cs.CV
cs.CV
cs.AI
Nuclear Medicine & Medical Imaging
09 Engineering
11 Medical and Health Sciences
Publication Status
Published
Date Publish Online
2016-10-29