Image Processing Device for Nucleated Red Blood Cell Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The existing image processing technologies face a high likelihood of false detection when searching for nucleated red blood cells in maternal blood due to the small number of target cells and the complexity of their detection from captured images.
Innovation Solution
An image processing device that extracts candidate nucleus pixels, forms connected pixel groups, and determines the presence of target cells by aggregating luminance gradients within specific regions without normalization, using machine-learned image features to reduce false detection.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Extent of automation
If traditional image processing methods are used to detect NRBCs, then the detection process can be automated, but the possibility of false detection increases due to the small number of target cells and complex background
Solution Approach 1:
The patent divides the detection process into multiple stages: initial pixel extraction based on color thresholds, connected component analysis to form candidate regions, and final classification using machine learning. This segmentation allows each stage to focus on specific features, improving both automation and reliability by systematically eliminating false positives at each step.
Solution Approach 2:
The patent performs preliminary filtering of pixels based on color characteristics before proceeding to more complex analysis. By pre-identifying pixels that match NRBC color profiles and grouping them into connected components, the system reduces the search space and eliminates obvious non-target elements early in the process, thereby improving detection reliability.
2Productivity
If simple pixel extraction methods are used, then the processing speed increases, but the detection precision decreases due to inability to distinguish target cells from background
Solution Approach 1:
The patent transitions from simple 2D pixel extraction to 3D feature space analysis by incorporating multiple attributes (color values, luminance gradients, texture features, and spatial relationships) for each candidate region. This dimensional expansion allows the machine learning classifier to distinguish target cells from background with high precision while maintaining processing efficiency through optimized feature calculation.
Solution Approach 2:
The patent dynamically adjusts detection parameters including color thresholds, luminance gradient sensitivity, and region grouping criteria based on the specific characteristics of each captured image. This adaptive parameter adjustment enables the system to maintain high processing speed while achieving accurate detection across varying image conditions.
3Measurement precision
If complex image processing algorithms are applied, then the detection accuracy improves, but the device complexity and computational load increase
Solution Approach 1:
The patent performs preliminary filtering and feature extraction to reduce the data volume before applying complex machine learning algorithms. By pre-identifying candidate regions and extracting only relevant features from these regions, the system minimizes the computational burden on the classification stage while maintaining high detection accuracy.
Solution Approach 2:
The patent replaces traditional rule-based image processing mechanisms with a machine learning-based classification system. This substitution allows the system to achieve high detection accuracy through learned patterns from training data, while the overall complexity is managed by using efficient feature extraction and a streamlined classification architecture.
Data Source
Figure 1
Figure 2
Figure 3~4
AI summary
The possibility of false detection is reduced when searching for a target cell by image processing from a captured image obtained by imaging a target cell having a nucleus, as compared with a case where the configuration of the present invention is not provided. A captured-image acquiring section 22 acquires a captured image obtained by imaging a sample including a target cell having a nucleus. A pixel extracting section 24 extracts pixels serving as a candidate for the nucleus from pixels included in the captured image. A connected-pixel-group extracting section 26 extracts a connected pixel group constituted of a predetermined number of adjacent connected pixels or more. A possibility determining section 40 determines a value indicating a possibility that the target cell is included within a region surrounding at least a portion of the connected pixel group. The value indicating the possibility is determined based on an image feature amount determined based on luminance distribution with respect to multiple different color components within the region. The value indicating the possibility is determined also based on a condition for an image feature amount machine-learned based on luminance distribution with respect to multiple different color components within a sample image of the target cell.