Tumor Segmentation Using Coarse-Fine Region Growing
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current tumor segmentation methods in medical imaging face challenges in accurately differentiating vital tumor tissue, edema, and necrotic tissue due to the lack of anatomical models, leading to information loss and leakage issues in boundary identification, especially in radiation therapy planning and treatment evaluation.
Innovation Solution
A medical imaging system comprising a workstation with a coarse segmenter, a fine segmenter, and an enclosed tissue identification module, which uses region growing algorithms to segment and classify tissue types by aggregating voxels based on maximum and minimum intensity values, preventing leakage and retaining edge information.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Speed
If geometric models or down sampling are used for fast computational measurement of tumor boundaries, then computational speed is improved, but edge information is lost
Solution Approach 1:
The patent divides the tumor segmentation process into two distinct stages: coarse segmentation that operates on down-sampled images for speed, and fine segmentation that processes original-resolution images for precision. This segmentation of the processing workflow allows both fast computation and high-quality edge detection to coexist by applying different processing levels to different aspects of the problem.
2Device complexity
If manual approaches based on geometric models are used for tumor boundary identification, then computational complexity is reduced, but segmentation accuracy deteriorates
Solution Approach 1:
The patent introduces an intermediary mechanism - the coarse segmentation result - that serves as a guide for the fine segmentation process. The coarse segmentation provides initial boundary estimates and region classifications that constrain and direct the more computationally intensive fine segmentation, reducing its search space while maintaining high accuracy in the final result.
3Speed
If growing algorithm with adaptive upper bound is used for tissue segmentation, then segmentation speed is improved, but leakage control becomes problematic
Solution Approach 1:
The patent performs preliminary coarse segmentation that identifies approximate tissue boundaries and classifications before conducting fine segmentation. This preliminary action establishes initial constraints and seed regions that guide the subsequent precise segmentation, preventing leakage by pre-defining plausible boundary locations based on intensity statistics from the coarse phase.
Data Source
AI summary
A medical imaging system (5) includes a workstation (20), a coarse segmenter (30), a fine segmenter (32), and an enclosed tissue identification module (34). The workstation (20) includes at least one input device (22) for receiving a selected location as a seed in a first contrasted tissue type and a display device (26) which displays a diagnostic image delineating a first segmented region of a first tissue type and a second segmented region of a second contrasted tissue type and identified regions which include regions fully enclosed by the first segmented region as a third tissue type. The coarse segmenter (30) grows a coarse segmented region of coarse voxels for each contrasted tissue type from the seed location based on a first growing algorithm and a growing fraction for each contrasted tissue type. The seed location for growing the second contrasted tissue type includes the first coarse segmented region and any fully enclosed coarse voxels, and each coarse voxel includes an aggregation of voxels and a maximum and a minimum of the voxel intensities. The fine segmenter (32) grows a segmented region of voxels for each contrasted tissue type from the seed location and bounded by the second coarse segmented region based on a second growing algorithm and a growing fraction for each contrasted tissue type initially set to the growing fraction for the corresponding region. The seed location for growing the second contrasted tissue type includes the first segmented region and any identified regions. The enclosed tissue identification module (34) identifies any regions of voxels fully enclosed by the first segmented region as being of the third tissue type. The coarse segmenter, the fine segmenter, and the enclosed tissue identification module are implemented by an electronic data processing device.