Sub-voxel Motion Correction for Phase-Contrast MRI
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image registration methods for blood flow studies, such as those using information theoretic approaches, face challenges in achieving sub-voxel registration accuracy due to interpolation artifacts, especially in images with significant gradient content or low contrast areas, which can render alignment results unreliable.
Innovation Solution
A continuous histogram method is introduced, where pixels are treated as 2D surface patches with continuous gray level distributions, updating a continuous set of joint histogram entries based on gradient information, rather than traditional interpolation methods that only update nearest neighbor entries, to estimate joint probability distributions and reduce mutual information estimation artifacts.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional interpolation methods are used for image registration, then the registration process is simple and fast, but sub-voxel registration accuracy cannot be achieved due to interpolation artifacts
Solution Approach 1:
The patent changes the fundamental parameter of how pixel values are represented - from discrete interpolated values to continuous probability distributions. By modeling pixel intensities as continuous random variables with probability density functions, the method eliminates interpolation artifacts while achieving sub-voxel accuracy. The continuous histogram entries represent probability distributions rather than single values, fundamentally changing the registration approach.
Solution Approach 2:
The patent introduces continuous histogram entries as an intermediary between discrete pixel values and registration metrics. These continuous histogram bins act as mediators that smooth out the discontinuities caused by traditional interpolation, allowing for more accurate mutual information calculation without the artifacts that plague conventional methods.
2Reliability
If discrete histogram entries are used, then the computation is simpler, but interpolation artifacts occur in images with gradient content or low contrast areas
Solution Approach 1:
The patent transitions from a discrete histogram representation to a continuous histogram representation by adding the dimension of continuity. Instead of updating only the nearest neighbor histogram bin for each pixel, the method updates a continuous range of histogram entries weighted by their probability, effectively moving from 0-dimensional (discrete bins) to continuous-dimensional representation.
3Measurement precision
If nearest neighbor interpolation is used, then the method is computationally efficient, but alignment results become unreliable in low contrast areas
Solution Approach 1:
The patent introduces dynamics into the histogram update process by making the update weights adaptive rather than static. The continuous histogram update uses gradient information to dynamically determine how much each pixel contributes to neighboring histogram bins, allowing the method to adapt to local image characteristics such as gradient content and contrast variations.
Data Source
AI summary
A method and system are described for image registration using an information theoretic approach that can be used for correcting motion in blood flow studies as well as other applications. The joint probability distribution between two MRI (or other modality) images is estimated where the interpolation method is referred to as a continuous histogram.