MRI Complex Image Denoising via Phase Intermediary
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Magnetic resonance imaging (MRI) grayscale images suffer from 'black floating' due to non-linear transformation from complex images, where noise follows a Rayleigh distribution, making it challenging to suppress this artifact using conventional smoothing filters.
Innovation Solution
A magnetic resonance imaging apparatus applies a weighted average filter to both real and imaginary parts of the complex image to generate a denoised phase image, which is then used to create a pseudo intensity image without non-linear transformation, allowing for noise reduction processing that maintains a Gaussian distribution, effectively reducing black floating.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of manufacture
If a non-linear transformation (polar coordinate transformation) is applied to generate a grayscale image from a complex image, then the image can be displayed in a conventional intensity format, but noise follows a Rayleigh distribution causing black floating artifact
Solution Approach 1:
The patent introduces a phase image as an intermediary component. Instead of directly transforming the complex image to grayscale (which causes black floating), the method separates the complex image into magnitude and phase components, processes the phase component separately, and then combines them to create the final grayscale image. This intermediary phase image allows noise reduction while maintaining the desired intensity representation.
Solution Approach 2:
The patent segments the complex image into two independent components: a magnitude image (real part) and a phase image (imaginary part). By processing these components separately - applying noise reduction to the magnitude image and preserving the phase image - the method avoids the black floating artifact that occurs when applying non-linear transformation to the complete complex image.
2Measurement precision
If a smoothing filter is applied to the grayscale image to denoise the image, then noise is reduced, but the black floating artifact cannot be resolved due to the bias in Rayleigh distribution
Solution Approach 1:
Instead of applying smoothing filters to the grayscale image after non-linear transformation (which cannot resolve black floating), the patent inverts the approach by working with the complex image components before transformation. By applying noise reduction to the magnitude image in the complex domain and preserving phase information, the method achieves noise reduction without introducing the Rayleigh distribution bias that causes black floating.
3Measurement precision
If noise amount estimation is performed using standard deviation of flat portion, then noise amount can be estimated, but accuracy degrades when flat portion region is small
Solution Approach 1:
The patent changes the parameter used for noise estimation from standard deviation of flat portion (which requires large flat regions) to a different metric that can accurately estimate noise even in regions with small flat portions. This parameter change allows reliable noise amount estimation across diverse image regions without requiring large flat areas.
Data Source
AI summary
According to one embodiment, a magnetic resonance imaging apparatus includes processing circuitry. The processing circuitry is configured to apply a filter to each of a first real-part image and a first imaginary-part image of a first complex image generated based on acquired magnetic resonance data and thereby generate a second complex image that includes a second real-part image and a second imaginary-part image. The processing circuitry is configured to generate a phase image denoised by the filter, the denoised phase image generated based on the second real-part image and the second imaginary-part image. The processing circuitry is configured to generate an intensity image related to an absolute value of the first complex image based on pixel values of the denoised phase image, the first real-part image, and the first imaginary-part image.