CoLIAGe MRI Feature Extraction for Pathology Differentiation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional methods for distinguishing subtly different pathologies, such as cancer subtypes, struggle to capture local orientation variations and pixel-scale image patterns due to their reliance on global texture features and domain-agnostic intensity variations, leading to difficulties in differentiating similar radiographic appearances.
Innovation Solution
The use of co-occurrence of local anisotropic gradient orientations (CoLIAGe) to capture higher-order patterns of local gradient tensors at a pixel level, computing a co-occurrence matrix on localized gradient tensors, and quantifying entropy features from MRI images, which are independent of absolute signal intensities and more robust to MRI drift.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional global texture analysis methods (GLCM, Gabor steerable features) are used to characterize pathologies, then global textural patterns can be captured, but local orientation variations and pixel-scale image patterns are lost due to averaging responses to a single global descriptor
Solution Approach 1:
The patent divides the image into multiple local neighborhoods or patches, computing texture features independently for each region. This segmentation allows preservation of local orientation variations and pixel-scale patterns while still capturing global textural characteristics through aggregation of local descriptors, resolving the contradiction between global pattern capture and local information retention
Solution Approach 2:
The patent applies different texture analysis operations to different local regions of the image, computing local binary patterns, gradient orientations, and co-occurrence matrices at the pixel level rather than globally. This local quality approach ensures that orientation-specific information is preserved in each neighborhood while maintaining overall diagnostic accuracy
2Measurement precision
If local binary patterns (LBP) are used to provide pixel-level response, then localized intensity variations can be captured, but the method becomes highly dependent on the radius parameter and fails to capture orientation information
Solution Approach 1:
The patent merges LBP with gradient orientation analysis and co-occurrence matrix computations. By combining these complementary approaches, the system captures both localized intensity variations (from LBP) and orientation information (from gradient analysis) without being overly dependent on the radius parameter, as the gradient-based features provide scale-invariant orientation cues
Solution Approach 2:
The patent creates a composite feature vector that integrates multiple texture descriptors including LBP, gradient orientations, and co-occurrence statistics. This composite approach leverages the strengths of each individual method while compensating for their weaknesses, reducing parameter sensitivity and enhancing overall diagnostic performance
3Adaptability or versatility
If conventional texture methods based on intensity variations are used, then domain-agnostic representations can be obtained, but histopathological differences manifested in differently oriented nuclei, lymphocytes, and glands are not reliably captured
Solution Approach 1:
The patent changes the analytical parameters from simple intensity variations to gradient orientations and co-occurrence statistics that are sensitive to structural arrangements. By computing features in multiple orientations and analyzing the co-occurrence of gradient directions, the method captures histopathological architecture (nuclei orientation, glandular patterns, lymphocyte distribution) while maintaining adaptability to different imaging domains through standardized mathematical operations
Data Source
AI summary
Methods, apparatus, and other embodiments associated with distinguishing disease phenotypes using co-occurrence of local anisotropic gradient orientations (CoLIAGe) are described. One example apparatus includes a set of logics that acquires a radiologic image (e.g., MRI image) of a region of tissue demonstrating disease pathology (e.g., cancer), computes a gradient orientation for a pixel in the MRI image, computes a significant orientation for the pixel based on the gradient orientation, constructs a feature vector that captures a discretized entropy distribution for the image based on the significant orientation, and classifies the phenotype of the disease pathology based on the feature vector. Embodiments of example apparatus may generate and display a heatmap of entropy values for the image. Example methods and apparatus may operate substantially in real-time. Example methods and apparatus may operate in two or three dimensions.