Multicontrast MRI Super-Resolution via Transformer-Empowered Multiscale Contextual Matching and Aggregation

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Biomedical Engineering

Background:

• Magnetic Resonance Imaging (MRI) offers diverse tissue contrasts, enabling multicontrast super-resolution (SR).
• Existing SR methods struggle with long-range dependencies and effective feature aggregation across contrasts.
• This limits the potential of multicontrast MRI for higher quality image reconstruction.

Purpose of the Study:

• To develop an advanced multicontrast MRI SR network addressing limitations of current approaches.
• To leverage transformer technology for improved feature extraction and aggregation in MRI SR.
• To enhance image fidelity and detail preservation in super-resolved MRI.

Main Methods:

• Introduced McMRSR++, a novel network utilizing transformers for multiscale feature matching and aggregation.
• Employed transformers to model long-range dependencies in reference and target MRI contrasts.
• Integrated a texture-preserving branch and contrastive loss for detailed texture restoration.

Main Results:

• McMRSR++ significantly outperformed state-of-the-art methods in quantitative metrics (PSNR, SSIM, RMSE).
• The method demonstrated superior restoration of anatomical structures and textural details.
• Experiments were validated on public and clinical in vivo MRI datasets.

Conclusions:

• McMRSR++ effectively addresses limitations in multicontrast MRI SR by using transformers.
• The proposed method enhances image quality and detail, showing potential for clinical applications.
• This advancement could lead to improved scan efficiency and diagnostic accuracy in MRI.