Transfer learning improves supervised image segmentation across imaging protocols

Publikation: Bidrag til tidsskriftTidsskriftartikelfagfællebedømt

Standard

Transfer learning improves supervised image segmentation across imaging protocols. / van Opbroek, Annegreet; Ikram, M. Arfan; Vernooij, Meike W.; de Bruijne, Marleen.

I: IEEE Transactions on Medical Imaging, Bind 34, Nr. 5, 2015, s. 1018-1030.

Publikation: Bidrag til tidsskriftTidsskriftartikelfagfællebedømt

Harvard

van Opbroek, A, Ikram, MA, Vernooij, MW & de Bruijne, M 2015, 'Transfer learning improves supervised image segmentation across imaging protocols', IEEE Transactions on Medical Imaging, bind 34, nr. 5, s. 1018-1030. https://doi.org/10.1109/TMI.2014.2366792

APA

van Opbroek, A., Ikram, M. A., Vernooij, M. W., & de Bruijne, M. (2015). Transfer learning improves supervised image segmentation across imaging protocols. IEEE Transactions on Medical Imaging, 34(5), 1018-1030. https://doi.org/10.1109/TMI.2014.2366792

Vancouver

van Opbroek A, Ikram MA, Vernooij MW, de Bruijne M. Transfer learning improves supervised image segmentation across imaging protocols. IEEE Transactions on Medical Imaging. 2015;34(5):1018-1030. https://doi.org/10.1109/TMI.2014.2366792

Author

van Opbroek, Annegreet ; Ikram, M. Arfan ; Vernooij, Meike W. ; de Bruijne, Marleen. / Transfer learning improves supervised image segmentation across imaging protocols. I: IEEE Transactions on Medical Imaging. 2015 ; Bind 34, Nr. 5. s. 1018-1030.

Bibtex

@article{07623de4ed6c4d29867aea9a1a639dba,
title = "Transfer learning improves supervised image segmentation across imaging protocols",
abstract = "The variation between images obtained with different scanners or different imaging protocols presents a major challenge in automatic segmentation of biomedical images. This variation especially hampers the application of otherwise successful supervised-learning techniques which, in order to perform well, often require a large amount of labeled training data that is exactly representative of the target data. We therefore propose to use transfer learning for image segmentation. Transfer-learning techniques can cope with differences in distributions between training and target data, and therefore may improve performance over supervised learning for segmentation across scanners and scan protocols. We present four transfer classifiers that can train a classification scheme with only a small amount of representative training data, in addition to a larger amount of other training data with slightly different characteristics. The performance of the four transfer classifiers was compared to that of standard supervised classification on two MRI brain-segmentation tasks with multi-site data: white matter, gray matter, and CSF segmentation; and white-matter- /MS-lesion segmentation. The experiments showed that when there is only a small amount of representative training data available, transfer learning can greatly outperform common supervised-learning approaches, minimizing classification errors by up to 60%.",
author = "{van Opbroek}, Annegreet and Ikram, {M. Arfan} and Vernooij, {Meike W.} and {de Bruijne}, Marleen",
year = "2015",
doi = "10.1109/TMI.2014.2366792",
language = "English",
volume = "34",
pages = "1018--1030",
journal = "I E E E Transactions on Medical Imaging",
issn = "0278-0062",
publisher = "Institute of Electrical and Electronics Engineers",
number = "5",

}

RIS

TY - JOUR

T1 - Transfer learning improves supervised image segmentation across imaging protocols

AU - van Opbroek, Annegreet

AU - Ikram, M. Arfan

AU - Vernooij, Meike W.

AU - de Bruijne, Marleen

PY - 2015

Y1 - 2015

N2 - The variation between images obtained with different scanners or different imaging protocols presents a major challenge in automatic segmentation of biomedical images. This variation especially hampers the application of otherwise successful supervised-learning techniques which, in order to perform well, often require a large amount of labeled training data that is exactly representative of the target data. We therefore propose to use transfer learning for image segmentation. Transfer-learning techniques can cope with differences in distributions between training and target data, and therefore may improve performance over supervised learning for segmentation across scanners and scan protocols. We present four transfer classifiers that can train a classification scheme with only a small amount of representative training data, in addition to a larger amount of other training data with slightly different characteristics. The performance of the four transfer classifiers was compared to that of standard supervised classification on two MRI brain-segmentation tasks with multi-site data: white matter, gray matter, and CSF segmentation; and white-matter- /MS-lesion segmentation. The experiments showed that when there is only a small amount of representative training data available, transfer learning can greatly outperform common supervised-learning approaches, minimizing classification errors by up to 60%.

AB - The variation between images obtained with different scanners or different imaging protocols presents a major challenge in automatic segmentation of biomedical images. This variation especially hampers the application of otherwise successful supervised-learning techniques which, in order to perform well, often require a large amount of labeled training data that is exactly representative of the target data. We therefore propose to use transfer learning for image segmentation. Transfer-learning techniques can cope with differences in distributions between training and target data, and therefore may improve performance over supervised learning for segmentation across scanners and scan protocols. We present four transfer classifiers that can train a classification scheme with only a small amount of representative training data, in addition to a larger amount of other training data with slightly different characteristics. The performance of the four transfer classifiers was compared to that of standard supervised classification on two MRI brain-segmentation tasks with multi-site data: white matter, gray matter, and CSF segmentation; and white-matter- /MS-lesion segmentation. The experiments showed that when there is only a small amount of representative training data available, transfer learning can greatly outperform common supervised-learning approaches, minimizing classification errors by up to 60%.

U2 - 10.1109/TMI.2014.2366792

DO - 10.1109/TMI.2014.2366792

M3 - Journal article

C2 - 25376036

VL - 34

SP - 1018

EP - 1030

JO - I E E E Transactions on Medical Imaging

JF - I E E E Transactions on Medical Imaging

SN - 0278-0062

IS - 5

ER -

ID: 127559526