Cell Image Discrimination via Multi-Path Normalization
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional image recognition technologies for cell discrimination in pathological diagnosis rely on features extracted from a single image, which are insufficient for accurately distinguishing cell types based on color shades, as they fail to differentiate between differences in cell shades and their surroundings or standard staining densities.
Innovation Solution
An image processing method that inputs a cell image and auxiliary information on color, generates multiple processed images through normalization processes using luminosity averages and variances specific to regions or cell types, and extracts feature quantities for accurate cell discrimination.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If feature quantities are extracted from a single input image using conventional image recognition technology, then the processing is simple and fast, but the cell discrimination accuracy is insufficient because it cannot differentiate between cell shade differences and standard staining density variations
Solution Approach 1:
The patent segments the image processing into multiple distinct stages: obtaining the input image, obtaining auxiliary information about standard staining density, generating multiple candidate processed images with different normalization parameters, evaluating each candidate, and selecting the optimal processing result. This segmentation allows complex accuracy-improving operations to be organized into manageable steps.
Solution Approach 2:
The patent performs preliminary action by obtaining auxiliary information about standard staining density before processing the cell image. This preliminary knowledge about expected staining characteristics is used to guide the normalization process, allowing the system to pre-adjust processing parameters based on known standards before actual cell discrimination takes place.
Solution Approach 3:
The patent implements feedback by evaluating each candidate processed image against the auxiliary information about standard staining density. The evaluation step provides feedback on how well each normalization approach preserves or restores expected staining characteristics, allowing the system to select the processing path that best matches known standards.
2Measurement precision
If multiple processed images are generated with different normalization parameters to improve cell discrimination accuracy, then the discrimination precision improves, but the processing time and computational load increase
Solution Approach 1:
The patent applies partial action by generating multiple candidate processed images with different normalization parameters only when necessary for accurate discrimination. The system evaluates each candidate and selects the most appropriate one, avoiding the need to process all possible variations. This partial exploration of the parameter space achieves sufficient accuracy without exhaustive computation.
Solution Approach 2:
The patent systematically changes normalization parameters to generate multiple candidate processed images. By varying parameters such as luminosity normalization factors and contrast adjustment values, the system explores different processing outcomes. This parameter exploration allows the selection of optimal processing conditions without requiring completely different processing approaches.
3Loss of information
If conventional feature extraction is used on a single image, then the processing is straightforward, but it cannot distinguish between differences in cell shades and differences between cell and surroundings
Solution Approach 1:
The patent introduces auxiliary information about standard staining density as an intermediary element between the raw cell image and the final discrimination result. This intermediary knowledge serves as a reference standard that helps distinguish whether observed features represent actual cell characteristics or merely variations in staining density, thereby preserving feature differentiation capability.
Solution Approach 2:
The patent adds another dimension to the analysis by incorporating auxiliary information about standard staining density as a separate data layer. Instead of relying solely on pixel intensity variations in the single image, the system compares image features against this additional dimensional reference, enabling differentiation between cell-related features and staining-related variations.
Data Source
AI summary
An image processing method includes image inputting, auxiliary information inputting, image processing, feature quantity extracting, and discriminating. The image inputting is inputting a cell image. The auxiliary information inputting is inputting auxiliary information on a color of the cell image. The image processing is generating multiple processed images by performing a different image-processing process on the cell image based on the auxiliary information. The feature quantity extracting is extracting a feature quantity of a discrimination target from each of the multiple processed images. The discriminating is discriminating the discrimination target in the cell image based on the feature quantity.