Self-supervised learning framework application for medical image analysis: a review and summary

Area of Science:

• Artificial Intelligence
• Medical Imaging
• Machine Learning

Background:

• Manual annotation of medical images is time-consuming, costly, and introduces bias.
• The rapid growth of medical data hinders supervised learning progress.
• Self-supervised learning (SSL) offers a solution by leveraging unlabeled data.

Purpose of the Study:

• To provide a comprehensive review of SSL methodologies in medical imaging.
• To systematically examine SSL applications from 2018 to September 2024.
• To guide medical professionals in integrating SSL into their research.

Main Methods:

• Systematic literature review of SSL in medical imaging.
• Analysis of studies across various modalities (CT, MRI, X-ray, Histology, Ultrasound).
• Categorization of applications including classification, segmentation, and performance enhancement.

Main Results:

• CT and MRI dominate SSL research, followed by X-ray, Histology, and Ultrasound.
• Contrastive learning is more prevalent than generative learning, except for CT and MRI.
• Segmentation tasks and MRI/Ultrasound classification show potential for further development.

Conclusions:

• SSL effectively addresses the limitations of manual annotation in medical imaging.
• The review highlights research trends and identifies areas for future SSL exploration.
• SSL holds significant promise for advancing AI in healthcare.