Prostate Cancer Tissue Classification via Deep Learning Nuclei Segmentation

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Deep learning networks for classifying prostate cancer tissue images face challenges due to significant variations in tissue image data caused by inter- and intra-patient tissue variations, malignancy grade variations, and tissue slide preparation variations, requiring very large and detailed annotated training data sets.

Innovation Solution

A method that uses a trained deep learning network to classify nuclei in prostate tissue images into categories such as stroma, benign tissue, prostatic intraepithelial neoplasia (PIN), and grading patterns, by capturing histological tissue image data from samples stained with light absorbent stains that differentiate stroma and nuclei, and applying a marked point process to identify nuclei and classify regions based on nuclear categories.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If deep learning networks are used to classify prostate cancer tissue images, then classification accuracy is improved, but very large and detailed annotated training data sets are required

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

Solution Approach 1:

The patent applies preliminary action by using unsupervised learning methods (BCD color decomposition, gland segmentation, nuclei identification) to pre-process and annotate training data automatically before supervised deep learning classification. This preliminary annotation of gland boundaries and nuclei locations reduces the manual annotation effort required for supervised training, addressing the contradiction between high classification accuracy and large training data requirements

Inventive Principle:
Principle #10Preliminary action

Solution Approach 2:

The patent introduces intermediary components (unsupervised learning algorithms for color decomposition, segmentation, and nuclei identification) that mediate between raw histological images and the supervised deep learning classifier. These intermediaries automatically generate preliminary annotations, reducing the burden of manual annotation while maintaining classification accuracy

Inventive Principle:
Principle #24Intermediary (Mediator)

2Reliability

If detailed annotated training data is collected to account for tissue variations, then classification reliability is improved, but the complexity of data annotation increases

Engineering Contradiction:
Improveclassification reliabilityVSAvoidannotation complexity
Core Design Contradiction:
ReliabilityVSDevice complexity

Solution Approach 1:

The patent applies self-service by implementing automated unsupervised learning algorithms (BCD color decomposition, gland segmentation, nuclei identification) that perform annotation tasks without extensive human intervention. These algorithms automatically adapt to tissue variations and generate consistent annotations, reducing annotation complexity while maintaining reliability through the detailed gland and nuclei level labeling

Inventive Principle:
Principle #25Self-service

Solution Approach 2:

The patent changes parameters by transitioning from direct manual annotation of cancer regions to automated unsupervised annotation of gland boundaries and nuclei locations. This parameter change in the annotation approach (from region-based to structure-based automated annotation) simplifies the annotation process while capturing detailed tissue heterogeneity for reliable classification

Inventive Principle:
Principle #35Parameter changes

3Measurement precision

If histological stains are used for visual analysis, then tissue differentiation is improved, but the stains are not ideal for automatic analysis

Engineering Contradiction:
Improvetissue differentiationVSAvoidautomatic analysis suitability
Core Design Contradiction:
Measurement precisionVSEase of operation

Solution Approach 1:

The patent replaces the mechanical/visual analysis approach with computational methods by applying BCD color decomposition and unsupervised learning algorithms to automatically analyze stained tissue images. This substitution maintains the tissue differentiation capability of histological stains while enabling automated quantification of gland and nuclei structures for computer-aided diagnosis

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

4Measurement precision

If manual annotation by pathologists is performed, then annotation accuracy is improved, but operator involvement and time consumption increase

Engineering Contradiction:
Improveannotation accuracyVSAvoidannotation time
Core Design Contradiction:
Measurement precisionVSLoss of time

Solution Approach 1:

The patent applies preliminary action by using unsupervised learning algorithms to pre-annotate gland boundaries and nuclei locations before supervised deep learning classification. This preliminary automated annotation reduces the time-consuming manual annotation work while maintaining accuracy through the detailed structure-based labeling that captures tissue heterogeneity

Inventive Principle:
Principle #10Preliminary action

Applied Scientific Principles

This section explains which scientific principles are used to turn an abstract innovation direction into a practical engineering solution.

Function Achieved in This Case

This method enables reliable training of deep learning networks on large amounts of prostate cancer data with accurate labeling of glandular structures, reducing operator involvement and allowing for adaptation to different stains and staining methods, while achieving precise classification of nuclei and regions in prostate tissue.

Implementation Method 1

Deep learning, and in particular deep convolutional neural networks, is emerging as a valuable tool in biomedical image analysis

Methodology Applied
Scientific EffectDeep learning:

Implementation Method 2

deep convolutional neural networks

Methodology Applied
Scientific EffectConvolutional neural networks:

Implementation Method 3

The method relies on a marked point process based on a Gibbs distribution that finds locations and approximate shape of both the stromal and epithelial nuclei

Methodology Applied
Scientific EffectMarked point process:

Implementation Method 4

The method decouples intensity from color information, and bases the decomposition only on the tissue absorption characteristics of each stain

Methodology Applied
Scientific EffectBlind color decomposition:

Implementation Method 5

bases the decomposition only on the tissue absorption characteristics of each stain

Methodology Applied
Scientific EffectAbsorption: Absorption (physical)

Implementation Method 6

one stain of said at least two stains being absorbed primarily by the stroma and another stain of said at least two stains being absorbed primarily by the nuclei

Methodology Applied
Scientific EffectLight absorption: Absorption (EM radiation)

Data Source

PatentEP3718046B1Prostate cancer tissue image classification with deep learning
Publication Date: 2025.04.09 CADESS MEDICAL AB
  • EP3718046B1 patent drawingFigure 1
  • EP3718046B1 patent drawingFigure 2
  • EP3718046B1 patent drawingFigure 3

AI summary

The method of the present invention classifies the nuclei in prostate tissue images with a trained deep learning network and uses said nuclear classification to classify regions, such as glandular regions, according to their malignancy grade. The method according to the present disclosure also trains a deep learning network to identify the category of each nucleus in prostate tissue image data, said category representing the malignancy grade of the tissue surrounding the nuclei. The method of the present disclosure automatically segments the glands and identifies the nuclei in a prostate tissue data set. Said segmented glands are assigned a category by at least one domain expert, and said category is then used to automatically assign a category to each nucleus corresponding to the category of said nucleus' surrounding tissue. A multitude of windows, each said window surrounding a nucleus, comprises the training data for the deep learning network.