3D Voxel Data Generation Using Probability Maps for Organ Segmentation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for generating three-dimensional shape data of organs, such as the heart, from tomographic images are limited in accuracy due to the influence of contrast media and the complexity of organ shapes, leading to incomplete extraction of regions like the right ventricular fluid and papillary muscles.
Innovation Solution
A shape data generation method and apparatus that transforms reference shape data to match designated regions in tomographic images, uses voxelization processing to identify boundaries, and employs probability maps and reaction-diffusion equations to extract accurate three-dimensional voxel data, overcoming limitations of brightness-based extraction methods.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If brightness-based extraction methods (Region Growing) are used to generate three-dimensional shape data, then the extraction process is simple and fast, but the extraction accuracy is insufficient due to influence of contrast media and complex organ shapes
Solution Approach 1:
The patent introduces an intermediary function (probability map) that maps brightness values to probabilities of belonging to the target organ. This probability map serves as a mediator between the simple brightness-based extraction and the complex organ boundaries, allowing accurate extraction while maintaining computational efficiency. The function is learned from training data and applied to new images to achieve both speed and accuracy.
Solution Approach 2:
The patent transforms the extraction problem from direct brightness thresholding to probability-based classification by changing the parameter space. Instead of using fixed brightness thresholds, the system uses a learned probability function that maps brightness values to membership probabilities, enabling accurate segmentation of complex organ shapes affected by contrast media.
2Device complexity
If conventional extraction methods are used, then the process is computationally simple, but regions like papillary muscles and right ventricular fluid cannot be extracted accurately
Solution Approach 1:
The patent performs preliminary actions by pre-processing the tomographic image data to enhance organ boundaries and reduce noise before the main extraction process. This includes applying filters and preprocessing operations that prepare the data for more accurate probability-based extraction, enabling better separation of complex regions like papillary muscles from the surrounding tissue.
3Use of energy by stationary object
If brightness thresholding is used for region extraction, then the method is computationally efficient, but it fails to accurately separate regions with similar brightness values affected by contrast media
Solution Approach 1:
The probability map function acts as an intermediary that transforms the unreliable direct brightness comparison into a reliable probability-based classification. This function is computed once and can be applied efficiently to multiple images, maintaining low computational cost while significantly improving extraction reliability for regions affected by contrast media.
Data Source
Figure 1~2
Figure 3
Figure 4
AI summary
A shape data generation method relating to this invention includes: first generating first three-dimensional voxel data that represents a target obj ect by using first tomographic images, in which a first region occupied by the target object is designated, among plural tomographic images; extracting, from the first tomographic images, brightness values of voxels included in the first region; second generating a function for calculating a probability that a voxel is included in the first region by using the extracted brightness values; calculating, for each voxel among voxels in a voxel space that includes the plural tomographic images, a probability by using a brightness value of the voxel and the function; and third generating second three-dimensional voxel data that represents the target object by using the first three-dimensional voxel data and probabilities calculated for the voxels in the voxel space.