Dynamic Transformer Based on Wavelet and Diffusion Prior Guidance for Cardiac Cine MRI Reconstruction.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Cardiovascular Diseases

Background:

• Cardiac magnetic resonance imaging (CMR) is crucial for diagnosing cardiovascular diseases due to its noninvasive nature and superior soft-tissue contrast.
• Accelerated cine MRI acquisition often involves undersampling, leading to artifacts like noise, aliasing, and loss of image detail.

Purpose of the Study:

• To develop an advanced framework for reconstructing high-quality cardiac cine MRI from undersampled data.
• To enhance detail recovery and temporal consistency in accelerated cine MRI while managing computational costs.

Main Methods:

• A novel wavelet-guided dynamic Transformer architecture was employed.
• Diffusion priors were integrated into a reduced latent feature space for high-frequency feature generation.
• Wavelet-domain decomposition and dynamic spatiotemporal modeling were utilized to capture structural information and temporal dependencies.

Main Results:

• The proposed method demonstrated superior reconstruction accuracy and temporal consistency compared to existing approaches.
• Effective detail recovery was achieved with minimal reverse sampling steps in the diffusion model.
• A favorable balance between computational efficiency and reconstruction performance was maintained.

Conclusions:

• The developed framework offers an effective and robust solution for accelerated cardiac cine MRI reconstruction.
• The combination of wavelet decomposition, diffusion priors, and dynamic Transformer modeling significantly improves image quality and temporal coherence.