WCCAN: Windowed Cross-contrast Attention Network for Multi-contrast Brain MR Image Super-resolution

Area of Science:

• Medical Imaging
• Image Reconstruction
• Artificial Intelligence

Background:

• Existing multi-contrast MRI super-resolution (MCMSR) methods often rely on computationally expensive spatial-domain fusion and global attention.
• These methods frequently overlook explicit high-frequency (HF) priors, limiting their effectiveness.
• There is a need for MCMSR methods that are computationally efficient and applicable to various contrast pairings.

Purpose of the Study:

• To develop a general reference-guided MCMSR framework with low computational cost.
• To enable application to any contrast pairing, moving beyond fixed clinical protocols.
• To address the limitations of existing MCMSR methods by incorporating explicit HF priors and reducing computational demands.

Main Methods:

• Introduction of a wavelet-guided HF prior modeling block for precise extraction and enhancement of anatomical details.
• Development of a triple cross-contrast fusion module utilizing windowed cross-contrast attention for efficient HF information transfer.
• Incorporation of a consistent feature fusion module with selective spatial adaptive modulation to minimize feature differences across contrasts.

Main Results:

• The proposed windowed cross-contrast attention network (WCCAN) framework consistently outperformed state-of-the-art MCMSR methods on IXI and M4Raw datasets.
• Demonstrated superior quantitative accuracy and visual fidelity in MRI super-resolution.
• Achieved significantly lower computational complexity and faster inference times compared to existing MCMSR approaches.

Conclusions:

• The WCCAN framework offers an efficient and accurate solution for MCMSR.
• It delivers superior reconstruction quality while reducing computational costs.
• The method shows promise for broader application in MRI super-resolution tasks.