Medical Image Region Selection via Precomputed Voxel Sorting
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for defining regions of interest in medical images, such as PET scans, are laborious and time-consuming, especially in 3D, due to the need for manual threshold adjustments and multiple user interactions, often resulting in slow processing speeds and inclusion of extraneous structures.
Innovation Solution
A method that sorts voxels according to a first variable, allowing users to select an initial voxel and define regions of interest with a single selection, using pre-processing to generate a list or hierarchy of voxels, enabling fast and accurate identification of regions without requiring extensive user interaction.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If manual threshold adjustment is used to define regions of interest, then the accuracy of region selection can be improved, but the time required for region selection increases significantly
Solution Approach 1:
The patent applies preliminary action by pre-calculating and storing intensity values for all possible bounding regions before the user makes a selection. The system processes and organizes region intensity data in advance, so when the user selects a bounding region, the intensity metrics are immediately available without requiring manual threshold adjustment or iterative analysis, thus resolving the contradiction between accuracy and time consumption.
2Manufacturing precision
If a user manually defines a bounding region to exclude extraneous structures, then the precision of region of interest can be improved, but the complexity of the operation increases
Solution Approach 1:
The system applies self-service by automatically calculating intensity values and identifying extraneous structures within the bounding region without requiring manual user intervention. The computer performs the analysis autonomously, using the user-defined bounding region as input and automatically providing the refined region of interest output, thus improving ease of operation while maintaining precision.
3Measurement precision
If multiple user interactions are required to adjust threshold and define regions, then the accuracy of lesion annotation can be improved, but the productivity of the system decreases
Solution Approach 1:
The system performs preliminary calculation of intensity values for all possible bounding regions and stores this data in advance. When a user defines a bounding region, the intensity metrics and lesion annotations are immediately available from pre-computed data, eliminating the need for iterative threshold adjustment and multiple user interactions, thus resolving the contradiction between accuracy and productivity.
4Device complexity
If a global threshold is applied to the entire image, then the simplicity of the algorithm can be improved, but the accuracy of region selection decreases due to inclusion of extraneous structures
Solution Approach 1:
The patent applies segmentation by dividing the image into multiple bounding regions and calculating intensity values for each region separately. This allows the system to apply thresholding locally to each region rather than globally, improving accuracy by excluding extraneous structures while maintaining algorithmic simplicity through systematic processing of divided regions.
Data Source
AI summary
In a method and apparatus for identifying regions of interest in medical images of a subject, in particular images captured by a medical imaging apparatus, such as a PET scanner, a list of voxels of the image is obtained, and sorted according to a first variable. A user-selection of an initial voxel in the image is then registered, and at least one voxel or group of voxels from the sorted list is selected as a region of interest, according to a property of the at least one voxel or group in relation to the user-selected initial voxel. The image may be pre-processed to generate the sorted list of voxels, following which the user selection of the initial voxel is registered.