Self-Supervised MR Image Reconstruction From Single Measurement.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Deep Learning

Background:

• Deep learning (DL) methods are increasingly used to accelerate magnetic resonance imaging (MRI).
• Training DL models for MRI reconstruction typically requires large amounts of paired data, which are difficult to acquire.
• Existing methods face challenges in data acquisition and artifact removal.

Purpose of the Study:

• To develop a self-supervised deep learning method for MRI reconstruction that eliminates the need for external training data.
• To improve the efficiency and practicality of DL-based MRI reconstruction.
• To address the data scarcity issue in DL-MRI training.

Main Methods:

• A single-image reconstruction approach inspired by Self2Self.
• Application of Bernoulli sampling to input images.
• Implementation of a drop strategy during training to mitigate artifacts in undersampled images.
• Integration of the physical principles governing MRI acquisition.

Main Results:

• The proposed self-supervised method successfully reconstructs MRI images without external training data.
• Experimental results indicate the method's effectiveness and good performance.
• The approach demonstrates potential for artifact reduction in undersampled MRI scans.

Conclusions:

• Self-supervised learning offers a viable solution for data-scarce DL-MRI reconstruction.
• The proposed method provides a practical and efficient alternative to traditional DL approaches.
• This technique has the potential to advance accelerated MRI acquisition.