Microscopy Instance Segmentation Using Voronoi Object Centers
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image processing methods for instance segmentation in microscopy require significant manual effort for training and annotation, and unsupervised methods provide only coarse divisions between foreground and background, lacking precision in distinguishing individual objects.
Innovation Solution
A method and system that uses Voronoi regions calculated from object centers in microscopy images to generate instance segmentation masks, reducing the need for manual training and annotation by combining object center determination and binary segmentation with Voronoi partitioning.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If machine-learned models for instance segmentation are used, then instance segmentation quality is improved, but manual annotation effort increases significantly
Solution Approach 1:
The patent segments the instance segmentation task into two independent parts: (1) binary segmentation to obtain foreground mask, and (2) Voronoi region assignment to differentiate instances. This eliminates the need for manual annotation while maintaining instance segmentation capability.
Solution Approach 2:
The patent introduces Voronoi regions as an intermediary mechanism between object centers and instance segmentation masks. By using Voronoi regions as a mediator, the system automatically assigns pixels to different instances based on proximity to object centers, replacing the need for manual instance-level annotation.
2Loss of time
If annotation-free segmentation methods are used, then manual effort is reduced, but only coarse division into foreground and background is achieved without instance discrimination
Solution Approach 1:
The patent adds a new dimension to segmentation by introducing instance identifiers through Voronoi region assignment. Instead of only spatial segmentation (2D), the method creates a third dimension of instance classification, enabling automatic differentiation of multiple objects of the same class without manual annotation.
Solution Approach 2:
The system uses automatically detected object centers to generate Voronoi regions that automatically assign pixels to different instances. The method is self-service in that it derives all necessary information (object centers, instance boundaries) from the image data itself without requiring external manual annotation.
3Measurement precision
If conventional machine-learned models are used for instance segmentation, then instance segmentation is achieved, but the model requires separate training and adaptation to current applications
Solution Approach 1:
The patent creates a universal instance segmentation approach that works across different applications without requiring application-specific training. By combining binary segmentation with Voronoi region assignment, the method provides a general-purpose solution that can be applied to various microscopy images regardless of the specific object type or application.
Solution Approach 2:
The patent replaces the complex machine-learned model system with a simpler geometric approach using Voronoi regions. This substitution eliminates the need for model training and adaptation while maintaining instance segmentation capability, making the system easier to implement and adapt to different applications.
Data Source
AI summary
A computer-implemented method for instance segmentation of at least one microscope image showing a plurality of objects, comprising: calculating positions of object centers of the objects in the microscope image; determining which image areas of the microscope image are covered by the objects; calculating Voronoi regions using the object centers as Voronoi sites; and determining an instance segmentation mask by separating the image areas covered by the objects into different instances using boundaries of the Voronoi regions.