Self-Supervised Visual Representation Learning for Medical Image Analysis: A Comprehensive Survey
Abstract: Deep learning has developed as a great tool for many computer vision or natural language processing tasks. However, supervised deep learning algorithms require a large amount of labelled data to achieve satisfactory performance. Self-supervised learning, a subcategory of unsupervised learning, circumvents the issue of the requirement of a large amount of data by learning representations from the data without labelled examples. Over the past few years, Self-supervised learning has been applied to various tasks to achieve performance at par with or surpassing the supervised counterparts in several tasks. However, the progress has been so rapid, that a comprehensive account of these developments is lacking. In this study, we attempt to present a review of those methods and show how the self-supervised learning paradigm evolved over the years. Additionally, we also present an exhaustive review of the self-supervised methods applied to medical image analysis. Furthermore, we also present an extensive compilation of the details of the datasets used in the different works and provide performance metrics of some notable works on image and video datasets.
Certifications: Survey Certification
Submission Length: Long submission (more than 12 pages of main content)
Changes Since Last Submission: As per the suggestions from Reviewer otY9,
- Sec. 3.2, 3.6 and 3.7 has been expanded with a few new studies added.
- Summary added at the end of Sec. 2, 3 and 4 to make the transitions smoother between sections.
- The language of the manuscript has been changed to improve readability and clarity.
As per the suggestions from Reviewer udZB,
- Sec 1.4 - Methodology has been added
- Declaration stating that we have not used ChatGPT for editing or writing any part of this review also added in Sec. 1.4
- The 'Dimension contrastive' section has been rightly renamed to 'Implicit Variance Regularization' and moved under the 'Non-contrastive Learning' section in Sec. 2.3.2.
- Figures 3, 4, and 5 added and depicts the illustrations of some foundational SSL frameworks.
- A short paragraph discussing the uses of SSL in medical image analysis and the type of downstream tasks has been added.
As per the suggestions from Reviewer uALV,
- Sec. 6 - Challenges and Limitations has been added
- A discussion on the future directions of research on SSL in medical image analysis
Changes in Camera Ready Version:
- Added missing references to K-Means algorithm, U-Net, CAGrad and CutMix papers.
- Modified the captions of Fig. 1 and 2 to better explain the abbreviations used in the figures.
- Previously unnumbered section "Organization of the survey" changed to a numbered section (Sec. 1.5) to avoid confusion in Sec. 1.4.
Assigned Action Editor: ~Jianbo_Jiao2
Submission Number: 2448
Loading