Self-supervised learning for MRI reconstruction: a review and new perspective.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Deep Learning

Background:

• Supervised deep learning (DL) for magnetic resonance imaging (MRI) reconstruction is limited by the need for extensive, fully sampled k-space data.
• Acquiring fully sampled k-space data is clinically impractical and costly, hindering the widespread adoption of supervised DL methods.
• Self-supervised learning (SSL) offers a viable alternative, enabling model training with readily available undersampled k-space data.

Purpose of the Study:

• To review recent advancements in self-supervised deep learning (DL) techniques for MRI reconstruction.
• To highlight the potential of SSL to overcome the data acquisition limitations inherent in supervised DL methods.
• To synthesize current research on SSL architectures and methodologies for improved MRI reconstruction.

Main Methods:

• A comprehensive literature review was conducted to identify and analyze recent developments in self-supervised DL for MRI reconstruction.
• The review focused on methods and architectures aimed at enhancing image quality, reducing scan times, and addressing data scarcity.
• Information was gathered from peer-reviewed publications and technical innovations in the field.

Main Results:

• Self-supervised DL techniques demonstrate significant potential for MRI reconstruction, effectively addressing data limitations.
• These methods can maintain high image quality while enabling faster MRI scans.
• SSL approaches offer a more feasible and scalable solution compared to traditional supervised methods.

Conclusions:

• Self-supervised DL is poised to transform MRI reconstruction by mitigating data dependency and improving efficiency.
• Key challenges remain, including ensuring robustness across different anatomies, standardizing validation protocols, and facilitating clinical integration.
• Future research should focus on hybrid approaches, domain-specific adaptations, and rigorous clinical validation to fully realize the potential of SSL in medical imaging.