CRL-NET: ACCELERATED MAGNETIC RESONANCE IMAGING RECONSTRUCTION THROUGH COIL REPRESENTATION LEARNING
Primary Area: unsupervised, self-supervised, semi-supervised, and supervised representation learning
Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.
Keywords: Magnetic Resonance Imaging, Medical Imaging, Representation learning, Computer Vision
Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2024/AuthorGuide.
Abstract: Magnetic Resonance Imaging(MRI) is a lengthy medical scan that stems from a long acquisition time. Its length is mainly due to the traditional sampling theorem, which defines a lower bound for sampling. However, it is still possible to accelerate the scan by using a different approach such as Compress Sensing(CS) or multi-coil Parallel Imaging(PI). These two complementary methods can be combined to achieve a faster scan with high-fidelity imaging. Recent advancements in Deep Learning (DL) have shown the potential to outperform traditional CS reconstruction techniques. This paper introduces CRL-Net, a novel Coil Representation Learning Network for accelerated multi-coil MRI reconstruction.The architecture of CRL-Net comprises a coil-wise encoder, devised to ascertain the distinctive representations of each coil. This is further complemented by a coil-attention layer, which synergistically assimilates inputs from both the sensitivity map estimations and the coil-wise encoder. Comprehensive evaluations of the CRL-Net, using the FastMRI benchmark for multi-coil datasets across knee and brain regions at both 4x and 8x acceleration, manifest significant advancements over the prevailing state-of-the-art methodologies. Such results elucidate the promising capability of CRL-Net in refining the accuracy and efficiency of MRI reconstructions.
Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors' identity.
No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.
Submission Number: 4785
Loading