Atlas Grid Gross Feature Recognition for Anatomical Images
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current transformation-based image analysis methods in medical imaging are sensitive to variations in image contrast and scanner parameters, making it difficult to accurately analyze anatomical images with varying intensity, volume, and shape, especially in clinical scenarios with heterogeneous patient populations and image quality.
Innovation Solution
The Gross Feature Recognition of Anatomical Images based on Atlas grid (GAIA) method, which uses anatomical parcellation maps to measure misregistration and intensity mismatch, allowing for the extraction of anatomical features by co-registering images to an atlas and ordering sub-regions based on intensity and texture features, rather than relying on direct anatomical information.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If transformation-based image analysis uses direct anatomical information (shape, volume, intensity), then anatomical features can be extracted, but the method becomes sensitive to variations in image contrast and scanner parameters
Solution Approach 1:
The patent changes the parameter basis from direct anatomical measurements (intensity, volume, shape) to ordinal rank orders of anatomical structures. By ranking structures based on their positions and characteristics rather than absolute measurements, the method becomes invariant to intensity scaling and scanner parameter variations, while still capturing anatomical feature information.
Solution Approach 2:
The patent creates an ordinal representation (copy) of the anatomical image data that preserves the relative relationships and hierarchical organization of structures without depending on absolute intensity values. This ordinal copy can be compared across different scans and scanners, enabling reliable feature extraction despite parameter variations.
2Measurement precision
If image quantification technologies are applied for group-based studies, then statistical analysis is improved, but the method becomes incompatible with individual image analysis in clinical practice
Solution Approach 1:
The patent segments the anatomical image into ordered sub-regions or parcels, creating a hierarchical structure that can be analyzed at multiple levels. This segmentation enables both individual case analysis (by examining the specific ordinal pattern of one patient) and group-based statistical analysis (by comparing patterns across multiple patients), thus bridging the gap between clinical and research applications.
Solution Approach 2:
The patent develops a multi-functional framework where the same ordinal representation and analysis methods can be applied to both individual diagnostic cases and group-based research studies. The method universally handles various anatomical structures and disease types through a unified approach, making it adaptable across different clinical and research scenarios.
3Ease of operation
If content-based image retrieval is implemented in PACS, then image searching capability is improved, but the difficulty of extracting features from stored images increases
Solution Approach 1:
The patent extracts only the essential ordinal features (rank orders of anatomical structures) from the full medical images, separating the critical diagnostic information from the complex raw data. This extracted ordinal representation serves as a compact feature set for CBIR, reducing the complexity of image comparison and retrieval while preserving the ability to distinguish different anatomical phenotypes.
Data Source
AI summary
Computer systems, computer-implemented methods, and non-transitory computer readable storage media for gross feature recognition including receiving an image comprising a plurality of image elements representing the region of interest of the subject. Gross feature recognition can further include co-registering the image to an atlas to segment the plurality of image elements into a plurality of sub-regions corresponding to structures in the atlas, where the structures in the atlas are ordered in a first rank order according to a predetermined feature. Further included can be ordering the plurality of sub-regions in a second rank order according to the predetermined feature. Further included can be identifying as gross features one or more of the plurality of sub-regions whose positions in the first rank order of the ordered sub-regions differ from positions in the second rank order of the corresponding ordered structures in the atlas.