Deep Learning Cell Morphology Analysis for Rare Cell Identification

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing cell identification methods, including manual microscopic observation and flow-type automatic hemocyte classification, struggle with accurately identifying cells with low emergence frequencies and require significant labor for training data generation, limiting analysis accuracy and generalization capability.

Innovation Solution

An image analysis method utilizing a deep learning algorithm with a neural network structure to calculate the probability of cell morphology classifications, enabling identification of cell types and abnormalities without manual observation, using training data generated from stained images.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If manual microscopic observation is used for cell identification, then examiner skill can be applied to identify cell types, but the number of preparations that can be observed per day is limited and identification of abnormal cells with low emergence frequency is difficult

Engineering Contradiction:
Improvecell identification accuracyVSAvoidnumber of preparations observed per day
Core Design Contradiction:
Measurement precisionVSProductivity

Solution Approach 1:

The patent replaces the manual mechanical observation system with an automated image analysis system using deep learning algorithms. The system captures images of cell preparations and uses neural networks to automatically identify cell types, replacing the examiner's manual microscopic observation with computational analysis, thereby increasing productivity while maintaining identification accuracy

Inventive Principle:
Principle #28Mechanics substitution (Replace mechanical system)

Solution Approach 2:

The system enables self-service by allowing the image analysis apparatus to autonomously perform cell identification without requiring continuous human intervention. The deep learning model automatically processes images, identifies cell types, and flags abnormal cells, making the system self-sufficient in performing the examination task

Inventive Principle:
Principle #25Self-service

2Productivity

If flow-type automatic hemocyte classification apparatus is used to increase the number of cell examinations, then productivity is improved, but information obtained is limited and identification of hemocytes with low emergence frequency is difficult

Engineering Contradiction:
Improvenumber of cell examinationsVSAvoidcell type identification accuracy
Core Design Contradiction:
ProductivityVSMeasurement precision

Solution Approach 1:

The patent replaces the flow-type automatic hemocyte classification system with an image-based deep learning analysis system. Instead of using flow cytometry principles, the system captures optical images of cell preparations and uses convolutional neural networks to identify cell types, enabling detection of abnormal cells with low emergence frequency that were missed by conventional automated systems

Inventive Principle:
Principle #28Mechanics substitution (Replace mechanical system)

Solution Approach 2:

The system changes the measurement parameters from flow cytometry-based physical properties to image-based morphological features. By analyzing visual characteristics such as cell size, shape, nuclear-to-cytoplasmic ratio, and staining patterns in captured images, the system achieves superior identification accuracy for rare abnormal cells compared to conventional automated classification methods

Inventive Principle:
Principle #35Parameter changes

3Extent of automation

If machine learning technique is used for cell identification, then automation is improved, but tremendous labor is required for creating training data and analysis accuracy is limited

Engineering Contradiction:
Improveautomation of cell identificationVSAvoidanalysis accuracy
Core Design Contradiction:
Extent of automationVSMeasurement precision

Solution Approach 1:

The system performs preliminary action by pre-processing captured images to enhance cell visibility and extract relevant features before feeding them to the deep learning model. The image processing steps including normalization, contrast enhancement, and feature extraction are performed in advance to optimize the input quality for the neural network, thereby improving analysis accuracy

Inventive Principle:
Principle #10Preliminary action

Solution Approach 2:

The patent introduces an intermediary layer between image capture and final classification by using a deep learning model with multiple processing layers. The convolutional neural network acts as an intermediary that automatically learns hierarchical features from raw images, transforming them into meaningful representations that improve classification accuracy without requiring manual feature engineering or extensive training data

Inventive Principle:
Principle #24Intermediary (Mediator)

Data Source

PatentEP3588368B1Image analysis method, apparatus, and deep learning algorithm generation method
Publication Date: 2026.03.18 SYSMEX CORP
  • EP3588368B1 patent drawingFigure 1
  • EP3588368B1 patent drawingFigure 2
  • EP3588368B1 patent drawingFigure 3

AI summary

Disclosed is an image analysis method including: inputting analysis data including information regarding an analysis target cell to a deep learning algorithm having a neural network structure; and analyzing an image by calculating, by use of the deep learning algorithm, a probability that the analysis target cell belongs to each of morphology classifications of a plurality of cells belonging to a predetermined cell group.