Spatial and Modal Optimal Transport for Fast Cross-Modal MRI Reconstruction

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Biomedical Engineering

Background:

• Multi-modal magnetic resonance imaging (MRI) is vital for disease diagnosis.
• Acquiring T2-weighted images (T2WIs) is time-consuming and susceptible to motion artifacts, hindering analysis.
• Existing image pre-processing methods require extensive parameter tuning.

Purpose of the Study:

• To develop an end-to-end deep learning framework to expedite T2WI acquisition using T1-weighted images (T1WIs).
• To mitigate spatial misalignment and motion artifacts in T2WI reconstruction.
• To improve the quality of T2WIs, especially at low sampling rates.

Main Methods:

• Utilized T1WIs as auxiliary data for expedited T2WI acquisition.
• Employed Optimal Transport (OT) for cross-modal synthesis and alignment of T1WIs to T2WIs.
• Implemented an alternating iterative framework combining reconstruction and cross-modal synthesis tasks.

Main Results:

• The proposed method effectively mitigates spatial misalignment.
• Experimental results on FastMRI and internal datasets demonstrate significant improvements in T2WI reconstruction quality.
• The framework shows effectiveness even at low sampling rates, enhancing diagnostic capabilities.

Conclusions:

• The deep learning framework successfully accelerates T2WI acquisition while improving image quality.
• The iterative optimization between reconstruction and synthesis enhances overall performance.
• This approach offers a promising solution for efficient and high-quality multi-modal MRI acquisition.