Two-Stage Segmentation for Overlapped Dot Counting in FISH Images
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In digital pathology, fluorescence in situ hybridization (FISH) images present challenges in detecting and separating overlapped dots due to two-dimensional projections of three-dimensional entities and low image contrast, leading to difficulties in accurate dot counting.
Innovation Solution
A two-stage segmentation framework is implemented, involving dot candidate seed determination through tophat transform, h-maxima suppression, and connected component analysis, followed by patch segmentation using local variance, adaptive dilation, and graph cuts, with classification stages utilizing shape and intensity-based features to accurately identify and count dots.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional dot counting methods are used on FISH images, then the process is simple, but overlapped dots cannot be separated and counted accurately
Solution Approach 1:
The patent divides the complex task of dot detection into multiple stages: (1) detecting dot candidate seeds using tophat transform and h-maxima suppression, (2) segmenting dot candidate patches using adaptive dilation and graph cuts, (3) extracting features, and (4) classifying candidates. This segmentation allows separated processing of overlapping structures while maintaining overall accuracy.
Solution Approach 2:
The patent transforms the two-dimensional image into a multi-dimensional feature space by extracting shape features (area, eccentricity, equivalent diameter) and intensity-based features (standard deviation, entropy, maximum intensity value). This dimensional transformation enables accurate differentiation and counting of overlapping dots that cannot be separated in the original 2D image space.
2Measurement precision
If image hard thresholding is applied to separate dots, then processing is simple, but artifacts with similar intensity range are generated
Solution Approach 1:
The patent changes the thresholding approach from fixed hard thresholding to adaptive thresholding based on local and mean variance. The adaptive dilation uses intensity values between [mean−variance, mean+variance], allowing dynamic adjustment to local image characteristics. This prevents generation of artifacts while maintaining accurate dot separation.
Solution Approach 2:
The patent applies different processing parameters to different regions of the image based on local variance characteristics. By determining local and mean variance for each dot candidate patch and using this information for adaptive dilation, the method adapts to local image quality variations, preventing artifact generation in low-contrast regions while maintaining separation accuracy.
3Productivity
If standard image processing is used, then processing speed is fast, but blurred dot boundaries with low contrast make separation challenging
Solution Approach 1:
The patent performs preliminary processing steps before main dot detection: applying tophat transform to enhance local maxima, performing h-maxima suppression to remove spurious peaks, and pre-segmenting dot candidate patches. These preliminary actions prepare the image data to highlight dot boundaries and reduce blur effects, enabling accurate separation without sacrificing processing speed.
Solution Approach 2:
The patent introduces an intermediary feature extraction and classification stage that mediates between the raw image data and final dot counting. By extracting shape features and intensity-based features as intermediaries, the system can accurately identify dot boundaries even in low-contrast, blurred regions, while maintaining efficient processing through automated classification.
Data Source
AI summary
Fluorescence in situ hybridization (FISH) enables the detection of specific DNA sequences in cell chromosomes by the use of selective staining. Due to the high sensitivity, FISH allows the use of multiple colors to detect multiple targets simultaneously. The target signals are represented as colored dots, and enumeration of these signals is called dot counting. Using a two-stage segmentation framework guarantees locating all potential dots including overlapped dots.