This work introduces a novel self-supervised k-space regularization method called PISCO (Parallel Imaging-inspired Self-Consistency) for improving neural implicit k-space (NIK) representations in motion-resolved abdominal MRI reconstruction.
The key highlights are:
PISCO leverages the inherent global k-space relationship without the need for explicit calibration data, enabling self-supervised refinement of the k-space predictions.
PISCO is seamlessly integrated into the training of NIK models, providing additional self-supervised regularization beyond the standard data consistency loss.
Experiments on simulated and in-vivo abdominal MRI data demonstrate that PISCO-NIK significantly outperforms state-of-the-art motion-resolved reconstruction methods in terms of spatial and temporal image quality, especially at higher acceleration factors.
The calibration-free and flexible design of PISCO makes it an attractive regularization method for further applications of k-space-based reconstruction techniques.
Overall, the proposed PISCO-NIK approach advances the field of motion-resolved abdominal MRI by enhancing the reconstruction quality through self-supervised k-space regularization, without the need for additional training data or calibration steps.
To Another Language
from source content
arxiv.org
สอบถามเพิ่มเติม