Medical Imaging Analysis Using Multi-Granularity Segmentation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current computational methods for medical imaging analysis lack the ability to dynamically control anatomical granularity, leading to low sensitivity in abnormality detection due to noisy information from high-resolution voxel-based analysis, and existing methods like isotropic spatial filtering result in information loss.
Innovation Solution
A computer-implemented method that segments imaging data into sub-regions corresponding to various structures at multiple levels of granularity, allowing for dynamic control of granularity based on anatomical structures, using a structure-template atlas to define pre-defined structures and calculate abnormality or risk factors at each level.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If voxel-based analysis is used to achieve maximum anatomical information, then measurement precision is improved, but reliability deteriorates due to noisy information from too many voxels
Solution Approach 1:
The patent segments the imaging data into a hierarchical structure with multiple granularity levels. At the highest level, the entire anatomical region is analyzed, then progressively divided into smaller sub-regions at lower levels. This segmentation allows the system to aggregate information from multiple voxels at each level, reducing noise while preserving anatomical detail. The hierarchical organization enables reliable abnormality detection by analyzing patterns across multiple levels rather than relying on individual noisy voxel data.
2Reliability
If isotropic spatial filtering is applied to reduce granularity, then reliability is improved by reducing noise, but loss of information increases
Solution Approach 1:
The patent implements local quality by applying different granularity levels to different anatomical sub-regions based on their specific characteristics. Each sub-region is analyzed at an appropriate level of detail rather than uniformly filtering the entire image. This allows the system to maintain high anatomical detail in regions where it is needed while applying noise reduction where appropriate, preserving information that uniform filtering would lose.
Solution Approach 2:
The patent adds a hierarchical dimension to the analysis by organizing voxels into multiple granularity levels. Instead of simply filtering in the spatial domain, the system creates a new dimensional structure where the same anatomical region can be analyzed at different levels of aggregation. This hierarchical dimension allows simultaneous preservation of fine anatomical detail and noise reduction by examining patterns across multiple levels.
3Device complexity
If single granularity level analysis is used to simplify the system, then device complexity is reduced, but adaptability deteriorates because human judgment dynamic control is not implemented
Solution Approach 1:
The patent implements dynamics by allowing the granularity level to be dynamically adjusted based on the analysis needs and anatomical structures being examined. The system can transition between different levels of the hierarchical structure during analysis, adapting to the specific requirements of different anatomical regions and abnormality types. This dynamic control mirrors human expert judgment, where the level of detail examined is flexibly adjusted based on the diagnostic context.
Solution Approach 2:
The hierarchical structure serves multiple functions simultaneously: it provides noise reduction through aggregation, preserves anatomical detail through finer segmentation, enables pattern recognition across scales, and allows adaptive analysis. This multi-functional design achieves high adaptability without proportionally increasing system complexity, as the same hierarchical framework supports multiple analytical objectives.
Data Source
AI summary
A method, computer system and computer readable storage medium for searching for one or more images having a region of interest similar to the region of a subject, including: receiving imaging data comprising a plurality of image elements of the region of interest of the subject; segmenting the imaging data of the region of interest of the subject into a plurality of sub-regions corresponding to various structures at a plurality of levels of granularity, the plurality of levels of granularity having a relationship such that a level of granularity has fewer structures at a lower level of granularity; and calculating at each of the plurality of levels of granularity an abnormality factor or risk factor for the segmented various structures of said region of interest, to provide a segmented said region of interest of said subject with at least one of said abnormality factor or risk factor associated therewith.