Blood Vessel Segmentation Using Multi-Model Mask Error Screening
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for segmenting blood vessels from angiography images using a single deep learning model suffer from low accuracy and numerous errors, necessitating improved techniques for generating more precise vessel segmentation results.
Innovation Solution
A method involving multiple blood vessel segmentation models is employed to generate candidate mask images, which are then evaluated for error levels based on various criteria such as pixel connectivity, blob analysis, diameter and brightness trends, and centerline length, allowing for the exclusion of erroneous images and the generation of a more accurate segmentation result.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a single deep learning model is used for blood vessel segmentation, then the device complexity is low, but the segmentation accuracy is insufficient and errors occur frequently
Solution Approach 1:
The patent applies segmentation by dividing the blood vessel segmentation task into multiple independent models, each specializing in different vessel types or regions. Multiple candidate mask images are generated by different segmentation models, which are then evaluated and combined to produce the final segmentation result, thereby improving accuracy while managing complexity through functional division
Solution Approach 2:
The patent merges multiple candidate mask images generated by different segmentation models into a single final segmentation result. The merging process includes evaluating error levels of each candidate and combining them through operations such as selecting the candidate with the lowest error level or synthesizing multiple candidates, thereby achieving higher accuracy than any single model could provide
2Measurement precision
If multiple blood vessel segmentation models are applied to generate candidate mask images, then the segmentation accuracy improves, but the processing time and computational complexity increase
Solution Approach 1:
The patent applies preliminary action by pre-evaluating candidate mask images using error level evaluation before final combination. Error levels are assessed based on predetermined criteria such as topology consistency, diameter trends, and brightness distribution, allowing the system to quickly identify and prioritize high-quality candidates without exhaustive processing of all possibilities
Solution Approach 2:
The patent changes parameters by adjusting the number of candidate mask images to be combined and the stringency of error level evaluation criteria. This allows flexible control over the balance between processing time and segmentation accuracy, enabling the system to adapt to different computational resource constraints while maintaining improved accuracy over single-model approaches
3Measurement precision
If candidate mask images are evaluated based on multiple criteria including topology, diameter, and brightness, then the error detection accuracy improves, but the evaluation complexity increases
Solution Approach 1:
The patent segments the error evaluation process into multiple independent criteria: topology evaluation (assessing connectivity and structure), diameter evaluation (checking vessel width consistency), and brightness evaluation (verifying intensity distribution). Each criterion operates independently and contributes to the overall error level assessment, making the complex evaluation process manageable and systematic
Solution Approach 2:
The patent creates a universal evaluation framework that uses the same error level assessment structure across different candidate mask images and different evaluation criteria. This multi-functional evaluation system can handle various types of errors (topological, dimensional, intensity-based) through a unified approach, reducing overall system complexity despite the multiple evaluation dimensions
Data Source
AI summary
A blood vessel image segmentating method according to an embodiment may comprise the steps of: generating a plurality of candidate mask images regarding a target blood vessel by applying a plurality of blood vessel segmentation models on a blood vessel image; evaluating an error level for each of the generated plurality of candidate mask images; and generating a target blood vessel segmentation result from the candidate mask images, on the basis of the evaluated error level.