Tagged-to-Cine MRI Sequence Synthesis via Light Spatial-Temporal Transformer.

Area of Science:

• Medical Imaging
• Biomedical Engineering
• Computational Anatomy

Background:

• Tagged magnetic resonance imaging (MRI) tracks internal tissue motion but requires additional cine MRI acquisition for analysis, increasing time and cost.
• Existing methods often process frames independently, missing complementary information across frames and struggling with tag fading, which impacts synthesis performance.

Purpose of the Study:

• To develop an efficient framework for tagged-to-cine MR sequence synthesis, reducing acquisition time and costs without compromising motion analysis.
• To leverage both spatial and temporal information effectively, even with limited data, to improve the synthesis of cine MRI from tagged MRI.

Main Methods:

• A novel framework utilizing a split-and-integral protocol for balanced spatial-temporal modeling.
• Implementation of the light spatial-temporal transformer (LiST^2) for exploiting local and global attention in motion sequences with efficient parameters.
• Adaptation of directional product relative position-time bias and shifted window for spatial-temporal correlation awareness and motion alignment, followed by a recurrent sliding fine-tuning (ReST) scheme for temporal consistency.

Main Results:

• The proposed framework demonstrates superior performance in tagged-to-cine MR synthesis compared to existing methods.
• Evaluation on paired tagged and cine MRI sequences confirmed the effectiveness of the spatial-temporal modeling and temporal consistency enhancement.

Conclusions:

• The developed framework offers an efficient solution for tagged-to-cine MR synthesis, potentially reducing medical imaging acquisition burdens.
• This approach preserves downstream motion analysis capabilities while improving efficiency and accuracy in synthesizing dynamic MRI sequences.