Biological Image Segmentation via Shape Ranking and SVM Hyperplane
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing digital image segmentation techniques for biological tissue samples are not robust enough to handle variations in histology, often resulting in over- or under-segmentation due to the use of weak segmentation algorithms that require significant adaptation and optimization for specific tissue types.
Innovation Solution
A computer-implemented method that uses support vectors to define a hyperplane in a vector space, mapping data points from digital images of biological tissue to segment cells based on their signed distance from the hyperplane, allowing for accurate segmentation by distinguishing between conforming and non-conforming shapes, and applying a weak segmentation algorithm like the watershed transform with optimized parameters.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If a weak segmentation algorithm is used, then the segmentation process is simple and fast, but the segmentation accuracy deteriorates due to over-segmentation or under-segmentation
Solution Approach 1:
The patent introduces an intermediary classification step between the weak segmentation algorithm and the final result. A classifier (such as a support vector machine) is trained to distinguish between correct and incorrect segmentations based on features extracted from the segmented regions. This intermediary component enables the system to use the fast weak segmentation algorithm while correcting its errors through the classifier, thus resolving the contradiction between speed and accuracy.
Solution Approach 2:
The system implements feedback by using the classifier to evaluate the output of the segmentation algorithm and provide correction signals. The classifier learns from training data what constitutes a correct segmentation and uses this knowledge to adjust or reject segmentations that do not meet the criteria. This feedback mechanism allows the system to maintain high speed while improving accuracy through iterative refinement.
2Device complexity
If a weak segmentation algorithm is used, then the device complexity is low, but the reliability of segmentation deteriorates across different tissue types
Solution Approach 1:
The patent creates a universal segmentation system that can handle multiple tissue types through the classifier. The classifier is trained on diverse training data representing different tissue types and segmentation qualities, enabling it to generalize across various histological contexts. This universal approach allows the simple weak segmentation algorithm to achieve reliable results across different tissue types without requiring tissue-specific optimization.
Solution Approach 2:
The system performs preliminary action by pre-training the classifier on extensive training data before actual segmentation tasks. During this offline training phase, the classifier learns the characteristics of correct segmentations across different tissue types. When performing actual segmentation, the pre-trained classifier can quickly evaluate and correct results without requiring complex real-time processing, thus maintaining low device complexity while achieving high reliability.
3Measurement precision
If segmentation parameters are optimized for one region, then the segmentation accuracy for that region improves, but the accuracy deteriorates in other regions with different histology
Solution Approach 1:
The patent employs parameter changes by using the classifier to dynamically adjust segmentation evaluation criteria based on the input image characteristics. Instead of using fixed parameters optimized for one region, the classifier learns to adapt its decision boundaries and feature weights based on the training data, which includes diverse tissue types. This allows the system to maintain high accuracy across different regions without requiring manual parameter re-optimization for each region.
Data Source
Figure 1
Figure 2A~2B
Figure 3~4A
AI summary
A method of segmenting a digital image of biological tissue includes accessing a ranking modelcalculated from training datarepresenting shapes of conforming and non-conforming biological unit exemplars. The ranking model may include support vectors defining a hyperplane in a vector space. The method further includes accessing image data representing the digital image, identifying a first shape and a set of second constituent shapes in the digital image, wherein the first shape comprises a union of the set of second constituent shapes, determining a rank of a first data point in the image data corresponding to the first shape and a rank of a second data point in the image data corresponding to the set of second constituent shapes into the vector space, and segmenting the digital image using the first shape or the set of second constituent shapes based on which data point has a greater respective rank.