Longitudinal Dynamic
MRI
Through Multisession Joint Reconstruction Incorporating Patient‐Specific Prior Image Information
Jingjia Chen, Hersh Chandarana, Daniel K. Sodickson, Li Feng ABSTRACT
Purpose
Serial MRI is common in clinical care and provides valuable subject‐specific information across imaging sessions that could be leveraged to improve imaging efficiency. However, current reconstruction methods typically process each session independently without using this rich temporal context. This work presents a novel concept of longitudinal dynamic MRI that incorporates patient‐specific prior images across multiple imaging sessions to enable higher acceleration than standard single‐session reconstruction.
Methods
The feasibility of longitudinal dynamic MRI reconstruction was demonstrated using the 4D Golden‐angle RAdial Sparse Parallel (GRASP) technique. Multi‐session GRASP datasets were concatenated into an extended dynamic series and reconstructed using a low‐rank subspace approach. Imaging experiments were performed on eight subjects across three separate sessions to evaluate reconstruction performance under different conditions.
Results
Compared with single‐session reconstruction, longitudinal 4D GRASP reconstruction enabled higher acceleration and achieved superior image quality across all subjects, even with up to a 5‐fold reduction in scan duration. Longitudinal reconstruction also demonstrated robustness to inter‐session variations in anatomy, body contour, imaging interval, and fat content without compromising reconstruction fidelity.
Conclusion
The proposed longitudinal dynamic MRI framework leverages shared patient‐specific information across imaging sessions to improve reconstruction quality and accelerate data acquisition. This work represents a promising imaging paradigm for applications involving repeated dynamic MRI scans, where imaging may become more efficient as additional patient‐specific data accumulate over time.