Spatially variant noise estimation in MRI: a homomorphic approach

Area of Science:

• Medical Imaging
• Signal Processing

Background:

• Accurate noise estimation is crucial for MRI post-processing.
• Existing methods often assume stationary noise, which is inadequate for modern MRI techniques like parallel reconstruction.
• Non-stationary noise requires advanced estimation methods.

Purpose of the Study:

• To develop a novel method for estimating non-stationary noise parameters from a single MRI magnitude image.
• To address the limitations of existing noise estimators that require multiple acquisitions or additional data.

Main Methods:

• Proposed a new method to estimate non-stationary noise parameters from a single magnitude MRI image.
• Utilized homomorphic separation to decompose spatially variant noise into stationary and low-frequency components.
• Applied low-pass filtering with Rician bias correction to estimate non-stationary noise variance.

Main Results:

• The proposed method accurately estimates non-stationary noise parameters from a single MRI acquisition.
• Demonstrated superior performance and lower error variance compared to state-of-the-art methods in both real and synthetic experiments.
• The method effectively handles non-stationary Rician noise, common in SENSE reconstructions.

Conclusions:

• The developed method provides a reliable way to estimate non-stationary noise in MRI without multiple acquisitions.
• This advancement is critical for improving the accuracy of post-processing algorithms in modern MRI.
• The technique offers a practical solution for noise characterization in advanced MRI scenarios.