Pathology Image Cell Clustering for Targeted Tumor Sequencing

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Current methods for genetic sequencing of tumors are inefficient due to cancer heterogeneity and the presence of healthy cells, leading to ineffective targeting and increased side effects, while high-resolution spatial transcriptomics are costly and not scalable.

Innovation Solution

An AI-based system for image processing of tissue specimens identifies cell groups with similar targets, allowing for targeted sampling and sequencing to understand genetic heterogeneity, predicting biomarkers, and optimizing treatment decisions.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Quantity of substance

If large parts of tumor tissue are sequenced to find genetic targets, then the coverage of cancer targets is improved, but the presence of healthy cells dilutes the target signal and reduces sequencing effectiveness

Engineering Contradiction:
Improveamount of tumor tissue sequencedVSAvoidtarget identification accuracy
Core Design Contradiction:
Quantity of substanceVSMeasurement precision

Solution Approach 1:

The patent segments the tumor tissue into multiple smaller regions of interest (ROIs) based on image analysis. Each ROI represents a distinct spatial area with potentially different genetic characteristics. By sequencing multiple smaller regions rather than one large region, the system captures tumor heterogeneity while avoiding healthy tissue contamination in each individual sequence, thus resolving the contradiction between quantity and precision.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The system applies different processing and sequencing strategies to different spatial regions of the tissue based on their identified characteristics. Each region is analyzed individually to determine its specific features, and sequencing is performed on selected regions that show high probability of containing relevant genetic targets. This local quality approach ensures that each sequencing effort is optimized for its specific region, improving overall target identification accuracy.

Inventive Principle:
Principle #3Local quality

2Measurement precision

If high-resolution spatial transcriptomics is used to identify different mutations in tumor regions, then the understanding of genetic heterogeneity is improved, but the cost increases and scalability decreases

Engineering Contradiction:
Improvegenetic heterogeneity resolutionVSAvoidcost and scalability
Core Design Contradiction:
Measurement precisionVSEase of manufacture

Solution Approach 1:

The system performs preliminary image-based analysis and region selection before sequencing. By using machine learning models to predict which regions are most likely to contain relevant genetic targets based on histological features, the system pre-screens the tissue and selects only the most promising regions for sequencing. This preliminary action reduces the number of regions that need expensive high-resolution spatial transcriptomics, making the overall process more cost-effective and scalable while maintaining high genetic heterogeneity resolution.

Inventive Principle:
Principle #10Preliminary action

Solution Approach 2:

The patent introduces an intermediate step between traditional histology and sequencing: machine learning-based region prediction. This intermediary system analyzes histological images to identify regions with high probability of containing specific genetic targets, serving as a bridge that guides subsequent sequencing efforts. This intermediary layer reduces the need for expensive comprehensive spatial transcriptomics by strategically selecting regions for detailed analysis, thereby improving cost-effectiveness and scalability.

Inventive Principle:
Principle #24Intermediary (Mediator)

3Productivity

If traditional sequencing methods are used on bulk tumor tissue, then the process is simple and scalable, but the heterogeneity of cancer causes loss of spatial information and reduces targeting effectiveness

Engineering Contradiction:
Improvesequencing throughputVSAvoidspatial information
Core Design Contradiction:
ProductivityVSLoss of information

Solution Approach 1:

The system segments the tissue into multiple discrete regions of interest based on image analysis, with each region representing a distinct spatial location. By processing and sequencing these segmented regions individually rather than as a bulk sample, the system preserves the spatial information of where each genetic sample originates. This segmentation approach maintains productivity by using automated image-based region identification while preventing information loss through spatial tracking of each sequenced region.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent adds a spatial dimension to the sequencing process by mapping genetic data back to their corresponding locations in the original tissue image. The system creates a spatial transcriptomics map where each sequenced region is positioned in its correct anatomical context. This dimensional addition allows the system to maintain high sequencing throughput while recovering and preserving spatial information that would otherwise be lost in bulk sequencing, enabling correlation of genetic heterogeneity with spatial distribution.

Inventive Principle:
Principle #17Another dimension (Dimensionality change)

Data Source

PatentEP4193299B1Systems and methods to process electronic images to provide image-based cell group targeting
Publication Date: 2026.04.01 PAIGE AI INC
  • EP4193299B1 patent drawingFigure 1A
  • EP4193299B1 patent drawingFigure 1B
  • EP4193299B1 patent drawingFigure 2

AI summary

Systems and methods are disclosed for grouping cells in a slide image that share a similar target, comprising receiving a digital pathology image corresponding to a tissue specimen, applying a trained machine learning system to the digital pathology image, the trained machine learning system being trained to predict at least one target difference across the tissue specimen, and determining, using the trained machine learning system, one or more predicted clusters, each of the predicted clusters corresponding to a subportion of the tissue specimen associated with a target.