Image Registration via Pixel Value Clustering and Region Division
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Image registration between perfusion images with different anatomical structures contrasted is challenging due to varying pixel values, leading to low matching degrees and inaccurate registration, especially when structures with high pixel values in one image appear low in another, causing loss of shape information necessary for registration.
Innovation Solution
An image registration method that acquires and clusters pixel value changes, divides images into regions based on pixel class information, and performs registration using representative values for each region, allowing for accurate alignment even when structures have contrasting pixel values.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If image registration is performed using pixel values from perfusion images with different anatomical structures contrasted, then registration can be performed between multiple phases, but the degree of matching becomes low and registration accuracy deteriorates when structures have different pixel values
Solution Approach 1:
The patent segments the image into multiple regions based on pixel value characteristics. By dividing the image into regions with similar pixel value patterns, the system can perform registration within each region separately, avoiding the problem of overall mismatch caused by contrasting anatomical structures with different pixel values across phases.
Solution Approach 2:
The patent applies different processing strategies to different regions of the image. Regions with high pixel values in one phase but low in another are handled differently from regions with consistent pixel values. This local differentiation preserves shape information in critical regions while accommodating variations in other areas.
2Reliability
If pixel value conversion is applied to entire images to solve contrast differences, then matching degree improves, but shape information of organs is lost
Solution Approach 1:
Instead of applying pixel value conversion to the entire image, the patent segments the image into multiple regions and applies conversion only where necessary. This selective approach maintains shape information in regions where it is critical while improving matching degree in regions where pixel value differences are the primary obstacle.
Solution Approach 2:
The patent applies different processing approaches to different regions: some regions undergo pixel value conversion to improve matching, while other regions preserve their original pixel values to maintain shape information. This localized quality adjustment resolves the contradiction between matching degree and shape preservation.
3Ease of operation
If standard image registration methods are used on perfusion images, then registration can be performed, but accurate matching fails when structures with high pixel values in one image appear low in another
Solution Approach 1:
The patent performs preliminary clustering of pixels based on their temporal pixel value changes before performing registration. This preliminary action groups pixels with similar contrast enhancement patterns together, creating a foundation for more accurate registration that accounts for phase-specific anatomical contrast differences.
Solution Approach 2:
The patent changes the parameter used for registration from absolute pixel values to temporal pixel value changes or ratios. This parameter transformation allows the system to capture anatomical structures regardless of whether they appear high or low in absolute intensity, improving matching accuracy while maintaining operational simplicity.
Data Source
AI summary
An image registration device includes: an image acquisition unit that acquires plural images captured in time series; a pixel value change acquisition unit that acquires a pixel value change at the same position of each of the images for plural positions of each image; a clustering unit that clusters the pixel value changes acquired for plural positions of each image into plural classes; a region division unit that divides each of the images into plural regions based on information of the class of each pixel of each image and a pixel value of each pixel of each image; and a registration processing unit that performs registration processing on each image based on information of plural regions of each image.