Cats or CAT scans: transfer learning from natural or medical image source data sets?

Veronika Cheplygina (Corresponding author)

Research output: Contribution to journalReview articleAcademicpeer-review

2 Citations (Scopus)
18 Downloads (Pure)

Abstract

Transfer learning is a widely used strategy in medical image analysis. Instead of only training a network with a limited amount of data from the target task of interest, we can first train the network with other, potentially larger source data sets, creating a more robust model. The source data sets do not have to be related to the target task. For a classification task in lung computed tomography (CT) images, we could use both head CT images and images of cats as the source. While head CT images appear more similar to lung CT images, the number and diversity of cat images might lead to a better model overall. In this survey, we review a number of articles that have studied similar comparisons. Although the answer to which strategy is best seems to be ‘it depends’, we discuss a number of research directions we need to take as a community to gain more understanding of this topic.
Original languageEnglish
Pages (from-to)21-27
Number of pages7
JournalCurrent Opinion in Biomedical Engineering
Volume9
DOIs
Publication statusPublished - Mar 2019

Fingerprint

learning
tomography
image analysis
train

Keywords

  • Deep learning
  • Medical imaging
  • Transfer learning

Cite this

@article{5806842d09b549cba9c0574cd04abebe,
title = "Cats or CAT scans: transfer learning from natural or medical image source data sets?",
abstract = "Transfer learning is a widely used strategy in medical image analysis. Instead of only training a network with a limited amount of data from the target task of interest, we can first train the network with other, potentially larger source data sets, creating a more robust model. The source data sets do not have to be related to the target task. For a classification task in lung computed tomography (CT) images, we could use both head CT images and images of cats as the source. While head CT images appear more similar to lung CT images, the number and diversity of cat images might lead to a better model overall. In this survey, we review a number of articles that have studied similar comparisons. Although the answer to which strategy is best seems to be ‘it depends’, we discuss a number of research directions we need to take as a community to gain more understanding of this topic.",
keywords = "Deep learning, Medical imaging, Transfer learning",
author = "Veronika Cheplygina",
year = "2019",
month = "3",
doi = "10.1016/j.cobme.2018.12.005",
language = "English",
volume = "9",
pages = "21--27",
journal = "Current Opinion in Biomedical Engineering",
issn = "2468-4511",
publisher = "Elsevier",

}

Cats or CAT scans: transfer learning from natural or medical image source data sets? / Cheplygina, Veronika (Corresponding author).

In: Current Opinion in Biomedical Engineering, Vol. 9, 03.2019, p. 21-27.

Research output: Contribution to journalReview articleAcademicpeer-review

TY - JOUR

T1 - Cats or CAT scans: transfer learning from natural or medical image source data sets?

AU - Cheplygina, Veronika

PY - 2019/3

Y1 - 2019/3

N2 - Transfer learning is a widely used strategy in medical image analysis. Instead of only training a network with a limited amount of data from the target task of interest, we can first train the network with other, potentially larger source data sets, creating a more robust model. The source data sets do not have to be related to the target task. For a classification task in lung computed tomography (CT) images, we could use both head CT images and images of cats as the source. While head CT images appear more similar to lung CT images, the number and diversity of cat images might lead to a better model overall. In this survey, we review a number of articles that have studied similar comparisons. Although the answer to which strategy is best seems to be ‘it depends’, we discuss a number of research directions we need to take as a community to gain more understanding of this topic.

AB - Transfer learning is a widely used strategy in medical image analysis. Instead of only training a network with a limited amount of data from the target task of interest, we can first train the network with other, potentially larger source data sets, creating a more robust model. The source data sets do not have to be related to the target task. For a classification task in lung computed tomography (CT) images, we could use both head CT images and images of cats as the source. While head CT images appear more similar to lung CT images, the number and diversity of cat images might lead to a better model overall. In this survey, we review a number of articles that have studied similar comparisons. Although the answer to which strategy is best seems to be ‘it depends’, we discuss a number of research directions we need to take as a community to gain more understanding of this topic.

KW - Deep learning

KW - Medical imaging

KW - Transfer learning

UR - http://www.scopus.com/inward/record.url?scp=85073763655&partnerID=8YFLogxK

U2 - 10.1016/j.cobme.2018.12.005

DO - 10.1016/j.cobme.2018.12.005

M3 - Review article

VL - 9

SP - 21

EP - 27

JO - Current Opinion in Biomedical Engineering

JF - Current Opinion in Biomedical Engineering

SN - 2468-4511

ER -