Automated Structure Labeling in H&E and IHC Tissue Images

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

The process of generating labeled data for biomedical imaging is labor-intensive and prone to errors due to the manual tracing of masks to delineate structures, which is inefficient and inaccurate for training machine learning systems.

Innovation Solution

A method involving staining tissue slides with hematoxylin and eosin (H&E) and immunohistochemistry (IHC) markers, followed by image preprocessing, registration, and automated region of interest (ROI) extraction, with manual and machine learning-based validation, to generate accurate labels for training data.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If manual tracing of masks is used to delineate structures, then labels can be generated, but the process becomes labor-intensive and error-prone

Engineering Contradiction:

Improvelabeling accuracyVSAvoidlabeling efficiency

Core Design Contradiction:

Measurement precisionVSProductivity

Solution Approach 1:

The patent replaces manual mechanical tracing with automated computational image processing. The system uses color deconvolution to separate H&E and IHC stain channels, automatically identifies epithelial-stromal boundaries through image analysis algorithms, and generates labels without human intervention, thereby eliminating the trade-off between accuracy and efficiency

Inventive Principle:

Principle #28Mechanics substitution (Replace mechanical system)

Solution Approach 2:

The system enables self-service labeling by allowing the tissue slide image and its color-separated channels to automatically identify and label structures of interest. The hematoxylin channel from H&E staining and the DAB channel from IHC staining self-organize to delineate boundaries without external manual input

Inventive Principle:

Principle #25Self-service

2Measurement precision

If multiple stains (H&E and IHC) are used to improve structure identification, then labeling accuracy improves, but image processing complexity increases

Engineering Contradiction:

Improvestructure identification accuracyVSAvoidimage processing complexity

Core Design Contradiction:

Measurement precisionVSDevice complexity

Solution Approach 1:

The patent applies segmentation by separating the multi-stain image into distinct color channels through deconvolution. The H&E image is separated into hematoxylin and eosin channels, while the IHC image is separated into hematoxylin and DAB channels. This segmentation simplifies processing by allowing independent analysis of each stain component rather than treating the multi-stain image as a complex whole

Inventive Principle:

Principle #1Segmentation

Solution Approach 2:

The hematoxylin channel serves as an intermediary that is common to both H&E and IHC staining methods. By using this shared channel for image registration and alignment, the system simplifies the integration of multiple stains, as the hematoxylin signal provides a consistent reference framework across different staining protocols

Inventive Principle:

Principle #24Intermediary (Mediator)

Data Source

PatentUS20250278951A1Systems and methods to label structures of interest in tissue slide images

Publication Date: 2025.09.04 MT SINAI SCHOOL OF MEDICINE

AI summary

Systems and methods to label structures of interest in tissue slide images are described.