Predicting 4D liver MRI for MR-guided interventions

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Biomedical Engineering

Background:

• Organ motion during interventions like radiation therapy presents a significant challenge.
• Current time-resolved volumetric magnetic resonance imaging (4D MRI) techniques lack the necessary speed and resolution for real-time guidance.

Purpose of the Study:

• To develop a novel deep learning approach for real-time, high-resolution 4D MRI with large fields of view for MR-guided interventions.
• To enable accurate prediction of 4D liver MRI with respiratory states from live 2D navigator MRI data.

Main Methods:

• A network-agnostic, end-to-end trainable deep learning formulation was developed.
• The method predicts 4D liver MRI from a live 2D navigator MRI, enabling both near real-time and retrospective reconstruction.
• Evaluated performance using mean target registration error (TRE) and visual comparison with state-of-the-art methods.

Main Results:

• Achieved near real-time (0.6s/volume) high-resolution (1.8mm isotropic) 4D MRI reconstruction.
• Demonstrated retrospective reconstruction with temporal resolution below 0.2s/volume.
• Reported a mean TRE of 1.19±0.74mm, below voxel size, with comparable quality to existing methods.
• Showed promising results with short training times (as low as 2 minutes).

Conclusions:

• The proposed deep learning formulation is highly effective for 4D MRI reconstruction.
• This approach significantly enhances the feasibility of 4D MRI for MR-guided interventions, improving motion management.
• The method offers flexibility for both real-time guidance and retrospective motion analysis in radiation therapy.