Self-supervised learning for medical image classification: a systematic review and implementation guidelines.

Area of Science:

• Artificial Intelligence
• Medical Imaging
• Computer Vision

Background:

• Deep learning models require extensive labeled data for medical image analysis, posing significant time and cost challenges.
• Self-supervised learning (SSL) emerges as a promising approach to leverage unlabeled medical datasets for model training.
• SSL can potentially enhance the development of robust medical imaging classification models.

Purpose of the Study:

• To systematically review and synthesize research on self-supervised learning applied to medical imaging classification.
• To provide a comprehensive overview of SSL strategies in the medical domain.
• To offer implementation guidelines for researchers in medical imaging classification.

Main Methods:

• Systematic literature review of papers published between 2012 and 2022.
• Searched databases: PubMed, Scopus, and ArXiv.
• Screened 412 studies, with 79 included for data extraction and analysis.

Main Results:

• Identified and described various self-supervised learning strategies applied to medical imaging.
• Synthesized collective knowledge from 79 selected studies.
• The review highlights the growing application and effectiveness of SSL in medical imaging classification.

Conclusions:

• Self-supervised learning is a viable and effective method for training medical imaging classification models without large labeled datasets.
• This review consolidates current knowledge and provides practical guidance for future research and implementation of SSL in medical AI.
• SSL holds significant potential to advance healthcare through improved medical image analysis.