Microscopy Cell Contour Segmentation for Low-Data Image Classification

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Deep neural network models for image analysis in medical microscopy require large training datasets and suffer from low computational efficiency, leading to high computational costs and susceptibility to human error in manual analysis.

Innovation Solution

A method that employs a contour recognition algorithm with an ideal cell radius, followed by image cropping, hole filling, binary conversion, and classification, utilizing mathematical calculations and machine learning to reduce data requirements and enhance computational efficiency.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If deep neural network models are used for image analysis, then analysis accuracy is improved, but computational cost and data requirements increase significantly

Engineering Contradiction:
Improveanalysis accuracyVSAvoidtraining data volume
Core Design Contradiction:
Measurement precisionVSQuantity of substance

Solution Approach 1:

The patent segments the image analysis task into distinct stages: initial contour detection using traditional algorithms, followed by targeted deep learning only for classification of detected cells. This segmentation allows traditional methods to handle the bulk of processing while deep learning enhances only the critical classification step, reducing overall data requirements.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent performs preliminary contour detection and cell isolation using traditional image processing algorithms before applying deep learning. This preliminary action filters and prepares the data, so that the deep learning model only needs to classify pre-processed cell images rather than learning from raw images, significantly reducing training data needs.

Inventive Principle:
Principle #10Preliminary action

2Measurement precision

If deep neural network models are used for image analysis, then analysis accuracy is improved, but computational efficiency decreases

Engineering Contradiction:
Improveanalysis accuracyVSAvoidcomputational efficiency
Core Design Contradiction:
Measurement precisionVSProductivity

Solution Approach 1:

The patent divides the analysis pipeline into efficient traditional processing stages (contour detection, cropping, hole filling) and a focused deep learning stage for classification only. This segmentation ensures that computationally intensive deep learning is applied only where necessary, maintaining high overall efficiency while achieving accurate results.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent performs preliminary image processing operations (contour extraction, cell cropping, binary conversion) using efficient traditional algorithms before deep learning classification. This preliminary action reduces the complexity of input data for the neural network, enabling faster and more efficient computation while maintaining accuracy.

Inventive Principle:
Principle #10Preliminary action

3Adaptability or versatility

If manual analysis is performed by diagnostic personnel, then flexibility and adaptability are maintained, but labor intensity increases and human errors occur

Engineering Contradiction:
Improvediagnostic flexibilityVSAvoidlabor intensity
Core Design Contradiction:
Adaptability or versatilityVSEase of operation

Solution Approach 1:

The patent implements an automated system where the diagnostic tool performs contour detection, cell extraction, and classification automatically without requiring manual intervention. The system serves itself by processing images through a pipeline of algorithms, eliminating labor-intensive manual analysis while maintaining diagnostic flexibility through configurable parameters.

Inventive Principle:
Principle #25Self-service

4Productivity

If traditional contour recognition algorithms are used with ideal cell radius, then computational efficiency is improved, but handling of variable cell sizes may be limited

Engineering Contradiction:
Improvecomputational efficiencyVSAvoidcell size variability handling
Core Design Contradiction:
ProductivityVSAdaptability or versatility

Solution Approach 1:

The patent uses dynamic contour detection that adapts to local image features rather than relying solely on a fixed ideal cell radius. The contour recognition algorithm dynamically adjusts to detect cell boundaries regardless of size variations, maintaining computational efficiency while handling variable cell sizes effectively.

Inventive Principle:
Principle #15Dynamics

Solution Approach 2:

The patent applies local quality by allowing the contour detection to adapt to local characteristics of each cell in the image. Instead of enforcing a uniform ideal radius, the algorithm detects contours based on local gradient information and intensity variations, enabling accurate detection of cells with varying sizes and shapes while maintaining overall computational efficiency.

Inventive Principle:
Principle #3Local quality

Data Source

PatentEP4668223A1A method for image processing and electronic device
Publication Date: 2025.12.24 NG WAI KWAN
  • EP4668223A1 patent drawingFigure 1
  • EP4668223A1 patent drawingFigure 2
  • EP4668223A1 patent drawingFigure 3~4

AI summary

The present application relates to a method for image processing and electronic device. The method includes: obtaining an original image; executing a predefined contour recognition algorithm with a given ideal cell radius value, to extract cell contour from the original image; according to the cell contour, cropping a plurality of cells from the original image; filling each of image holes with a predefined pixel value to generate a filed image; converting the filled image into a corresponding binary image; according to the binary image, identifying a plurality of remaining cells in the filled image; and classifying each of cells in the original image.