The PHU-NET: A robust phase unwrapping method for MRI based on deep learning.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Image Processing

Background:

• Magnetic Resonance (MR) phase imaging is crucial for various applications.
• Traditional phase unwrapping methods face challenges with robustness and efficiency.
• Improving MR phase unwrapping is essential for enhanced diagnostic capabilities.

Purpose of the Study:

• To develop a deep learning-based approach for MR image phase unwrapping.
• To enhance the robustness and computational efficiency of existing methods.
• To introduce a novel network, PHU-NET, for improved MR phase unwrapping.

Main Methods:

• A deep learning network, PHU-NET, was designed for MR phase unwrapping.
• A novel training data generation method was proposed to simulate wrapping patterns.
• Explicit estimation of wrapping boundaries and counts was used for network training.
• Quantitative evaluation was performed on simulated and real human body MR phase images with varying SNR.

Main Results:

• The proposed method demonstrated superior performance on simulated data, even at extremely low SNR.
• PHU-NET achieved reduced residual wrapping in human body MR images.
• The method proved effective in the presence of severe anatomical discontinuity.
• Significant improvements in computational efficiency were observed compared to traditional methods.

Conclusions:

• A robust and computationally efficient MR phase unwrapping method using deep learning was developed.
• PHU-NET shows promising performance for applications relying on MR phase information.
• This deep learning approach offers a significant advancement over traditional MR phase unwrapping techniques.