Development of Voxel-By-Voxel-Based Synthetic CT Generation Method Using a Flexible Ute Sequence (FUSE) In a Porcine Model
Abstract
Purpose
In conventional MRI, bone and air appear as signal voids due to signal decay, making them difficult to distinguish. This remains a major limitation for synthetic CT (sCT) generation in MRI-based radiotherapy. Although AI-based approaches have shown promise, they often perform poorly in complex anatomical regions particularly at bone–air interfaces due to limited generalizability and reliance on training datasets. This study investigates a novel Flexible UTE (FUSE) sequence to improve bone–air characterization and demonstrates its feasibility within a voxel-based sCT generation framework.
Methods
The FUSE sequence is a flexible ultrashort echo time acquisition that supports non-Cartesian k-space sampling, long-T2 tissue signal suppression, and off-resonance correction, with offline reconstruction. Sequence performance was evaluated using a porcine head model, which closely resembles human cranial anatomy and contains a complex mixture of bone, air, and soft tissue. A high-resolution CT scan was acquired as reference. FUSE images were obtained alongside conventional MRI sequences (T1, T2, and Dixon) to enable multiparametric voxel-wise analysis.
Results
T1- and T2-weighted images confirmed that the porcine model provides rich anatomical detail with complex bone–air–soft tissue interfaces, making it suitable for UTE investigations. Multiple bony and soft-tissue landmarks were consistently identified across imaging modalities, including several craniofacial anatomical landmarks. The FUSE sequence was implemented and executed on a 3T MRI (Siemens, Erlangen, Germany) system and demonstrated sensitivity to bone–air interfaces not visible on conventional MRI. Further optimization is required to improve signal-to-noise ratio and reduce motion-related artifacts. The combined multiparametric dataset provides a strong foundation for voxel-wise tissue characterization toward sCT generation.
Conclusion
The porcine head model is a realistic platform for sCT research, particularly for complex bone–air interfaces. These results extend prior phantom-based FUSE findings to anatomical settings. Future work will focus on optimizing FUSE and integrating multiparametric MRI for voxel-based sCT generation.