Complex Number Deep Networks for MRI Denoising and Reconstruction
Hyeyeon Chung Fuyao Li Yiqing Wang
hyeyeon.chung@duke.edu fuyao.li@duke.edu yq.wang@duke.edu
| Paper PDF |
|
This project investigates the utilization of complex-number deep networks for denoising and reconstructing MRI images. By leveraging the properties of complex numbers, which are inherent in MRI data, our models aim to provide more accurate and efficient image reconstruction compared to traditional real-number based methods. Our experimental focus is on three primary challenges: the elimination of Rician noise, the deblurring of motion artifacts, and the acceleration of imaging processes using under-sampled data. Contrary to our initial hypothesis, we observed that the real-number network outperformed the complex-number network. Moreover, replacing RadialNorm with BatchNorm in the complex-number network resulted in enhanced performance.
|
|
|
| Paper: |
| Code and Data: |