Object Segmentation in Medical Images Using Defined Regions
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing medical imaging segmentation methods face challenges in achieving a balance between speed and accuracy, particularly in three-dimensional contexts, due to computational complexity and the need for full volume grid scans, which increase cost and time, while interactive approaches require manual effort and lack contextual enforcement.
Innovation Solution
A method and device for segmenting objects in medical images by defining a segmentation region, such as a bounding box, within a source image, allowing for fast and accurate segmentation of selected elements using predefined models and fine-grained grid processing.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If machine learning methods are used for segmentation, then segmentation accuracy is improved, but computational time increases causing lag
Solution Approach 1:
The patent divides the full volume grid scan into smaller regions by first determining a segmentation region of interest around the selected element. This spatial segmentation allows the system to process only the relevant portion of the image, reducing computational time while maintaining segmentation accuracy for the target element.
Solution Approach 2:
The patent applies different processing strategies to different regions: a full grid scan is performed only within the determined segmentation region containing the selected element, while other regions are skipped. This local quality approach optimizes computational resources by applying high-accuracy segmentation only where needed.
2Reliability
If a full volume grid scan is used for segmentation, then segmentation completeness is improved, but computational cost and time increase
Solution Approach 1:
The patent extracts only the necessary portion of the data for segmentation by determining a segmentation region that contains the selected element and processing only that region. This extraction approach maintains segmentation completeness for the target element while significantly reducing overall computational cost.
Solution Approach 2:
The patent applies partial action by performing a full grid scan only within the determined segmentation region rather than the entire volume. This partial scan provides sufficient coverage for the selected element without the excessive computational burden of scanning the complete volume.
3Measurement precision
If transformer architectures with pre-trained autoencoders are used, then segmentation accuracy is improved, but computational complexity increases quadratically
Solution Approach 1:
The patent segments the input image into a smaller segmentation region of interest before applying the transformer architecture. This reduces the input size to the complex model, thereby reducing quadratic computational complexity while maintaining the accuracy benefits of transformer-based segmentation.
Data Source
AI summary
Method and device for segmenting an object in a source image. A source image showing details of the object is provided and an element of the object is selected. A segmentation region in the source image comprising the selected element is determined. A segmentation mask is prepared by segmenting the determined segmentation region, wherein at least the selected element or its components are segmented. The segmentation mask is then output.