Histopathology Cell Classification Across Scanner and Staining Variability

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing automated systems for pathological cell and tissue examination using artificial intelligence struggle to generalize across different scanners and staining systems, leading to inconsistent quality and reproducibility, and are time-consuming.

Innovation Solution

A method for training an artificial intelligence system by manually marking cell centers in digital images, using statistical distributions to create training data, and employing a neural network architecture that separates cell and context paths for improved detection and classification, allowing the system to handle scanner and staining variability.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Productivity

If automated image recognition using artificial intelligence systems is used to analyze heterogeneous tissue image data from different scanners, then productivity is improved, but measurement precision deteriorates due to scanner variability affecting generalization

Engineering Contradiction:
Improveautomated analysis throughputVSAvoiddetection accuracy across different scanners
Core Design Contradiction:
ProductivityVSMeasurement precision

Solution Approach 1:

The patent transforms fixed规则的 detection into adaptive learning by changing the parameters from manually defined thresholds to statistically learned distributions. The system learns optimal detection parameters from training data, allowing it to adapt to different scanner characteristics and staining variations, thereby maintaining high detection accuracy across heterogeneous image sources

Inventive Principle:
Principle #35Parameter changes

Solution Approach 2:

The patent creates a statistical model (copy) of cell distribution patterns from training data that can be applied across different scanners. Instead of requiring perfect reproduction of absolute image characteristics, the system copies the relative statistical relationships and patterns, making the detection robust to scanner-specific variations

Inventive Principle:
Principle #26Copying

2Device complexity

If conventional computer vision methods with manually defined rules are used, then device complexity is reduced, but manufacturing precision deteriorates due to insufficient quality standard consistency

Engineering Contradiction:
Improvesystem simplicityVSAvoidquality standard consistency
Core Design Contradiction:
Device complexityVSManufacturing precision

Solution Approach 1:

The system performs self-training by automatically learning from provided training images without requiring manual programming of detection rules. The artificial intelligence system self-optimizes its detection parameters and patterns, eliminating the need for complex manual rule configuration while achieving consistent high-quality results across different applications

Inventive Principle:
Principle #25Self-service

Solution Approach 2:

The patent applies preliminary training action by pre-training the system on labeled training data before actual detection. This preliminary learning phase establishes the statistical foundations and detection patterns that enable consistent performance in subsequent applications, avoiding the need for complex real-time rule adjustments

Inventive Principle:
Principle #10Preliminary action

3Measurement precision

If manual analysis of slides is performed, then measurement precision is maintained through expert judgment, but loss of time increases due to time-consuming and repetitive work

Engineering Contradiction:
Improvediagnosis accuracyVSAvoidanalysis time
Core Design Contradiction:
Measurement precisionVSLoss of time

Solution Approach 1:

The patent replaces the mechanical process of manual visual inspection with an automated artificial intelligence system that processes images computationally. This substitution maintains diagnostic accuracy by learning from expert-labeled data while eliminating the time constraints and human fatigue associated with manual analysis

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

Solution Approach 2:

The system enables continuous automated analysis without the interruptions, fatigue, and variability inherent in manual processes. The artificial intelligence system can process images continuously at high speed while maintaining consistent performance, eliminating the repetitive nature of manual work

Inventive Principle:
Principle #20Continuity of useful action

Data Source

PatentEP4246468B1Method for training an artificial intelligence system, method for recognizing and classifying cells for pathological cell and tissue examination, classifying system, computer-readable medium
Publication Date: 2025.12.10 MINDPEAK GMBH
  • EP4246468B1 patent drawingFigure 1~2
  • EP4246468B1 patent drawingFigure 3~4
  • EP4246468B1 patent drawingFigure 5

AI summary

Analysing digital images using artificial intelligence systems for pathological analysis currently does not generalize well over a large number of digital images with variable image characteristics, different staining systems and different laboratory processes. The problem is solved by a method for training an artificial intelligence system (1) and for recognizing and classifying cells for histopathological tissue examination, under use of the artificial intelligence system (1), wherein in a learning stage, the following steps are performed: g) manual marking of cells (13, 14) to be detected by the artificial intelligence system (1) in digital images of histological tissue sections, h) each marking is assigned to either one of at least one tumor cell class or one of at least one non-tumor cell class, i) a number of marked digital images according to steps a) and b) are entered into the artificial intelligence system (1) for deep learning purposes, j) the artificial intelligence system (1) learns characteristics of the at least one tumor cell class and the at least one non-tumor cell class from the marked digital images to establish a model for cell detection and classification, in an analysis stage digital images are obtained and cells are detected and classified according to the model.