Simultaneous Phase Denoising And Unwrapping Of Mr Images Using A New Deep Learning Method

The existing approach for unwrapping noisy MR images involves a two-step process: denoising followed by phase unwrapping. However, such a sequential approach could introduce errors, particularly in regions with high noise and abrupt phase transitions. To address this issue, this paper introduces a novel deep learning approach that simultaneously performs denoising and phase unwrapping. The proposed architecture features a unified encoder and two decoders connected through skip connections. The effectiveness of the proposed method is assessed using two distinct datasets: (i) simulated data based on a Gaussian mixture model with varying noise levels and a dynamic range of phase values and (ii) real brain MR image data. The results of the proposed method are compared with existing phase unwrapping methods like Laplacian, SEGUE, UNet, and UNet++. The proposed method has consistently outperformed these existing methods, even at a Signal to Noise Ratio (SNR) as low as -5 dB.