Whole-Slide Image Quality Control via Attention-Based Tile Selection

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Current machine learning models for whole-slide image analysis lack interpretability and may generate accurate results for irrelevant reasons, leading to inefficiencies in diagnostic processes.

Innovation Solution

Subdivide whole-slide images into tiles, apply a trained machine learning model to generate predictions, determine attention scores for each tile, and select a subset based on these scores to assess biological relevance, generating pass/fail indications.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If machine learning models process all image tiles in whole-slide images to generate diagnostic predictions, then prediction accuracy is improved, but computational time and resources increase significantly

Engineering Contradiction:
Improveprediction accuracyVSAvoidcomputational time
Core Design Contradiction:
Measurement precisionVSLoss of time

Solution Approach 1:

The patent divides the whole-slide image into multiple tiles and further segments these tiles into sub-tiles. This segmentation allows the model to process smaller units individually, enabling selective attention to relevant regions. The attention mechanism then weights these sub-tiles to focus computational resources on diagnostically important areas rather than processing all tiles uniformly, thus reducing overall computational time while maintaining prediction accuracy.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent applies local quality by using an attention mechanism that assigns different weights to different sub-tiles based on their diagnostic relevance. Regions with higher attention scores (indicating higher biological relevance) are processed with greater detail and weight, while less relevant regions receive reduced processing. This allows the system to allocate computational resources efficiently, focusing on critical areas that most impact prediction accuracy.

Inventive Principle:
Principle #3Local quality

2Measurement precision

If machine learning models analyze all regions of whole-slide images, then diagnostic accuracy is improved, but interpretability of model decisions deteriorates

Engineering Contradiction:
Improvediagnostic accuracyVSAvoidmodel interpretability
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

By segmenting the image into tiles and sub-tiles, the patent creates discrete, manageable units that can be individually evaluated. The attention mechanism then identifies and highlights specific sub-tiles that contribute most to the prediction, making it easier to interpret which regions drove the diagnostic decision. This segmentation transforms a black-box analysis of the entire slide into an interpretable set of focused regional assessments.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The attention mechanism serves as an intermediary between the raw image data and the final prediction. It generates attention scores that act as interpretable intermediaries, indicating the relative importance of each sub-tile. These attention scores provide a bridge that allows clinicians to understand which regions influenced the prediction, enhancing model interpretability without sacrificing diagnostic accuracy.

Inventive Principle:
Principle #24Intermediary (Mediator)

3Manufacturing precision

If machine learning models process high-resolution whole-slide images in detail, then diagnostic precision is improved, but computational resources and processing time increase

Engineering Contradiction:
Improvediagnostic precisionVSAvoidcomputational resources
Core Design Contradiction:
Manufacturing precisionVSUse of energy by moving object

Solution Approach 1:

The patent segments high-resolution whole-slide images into multiple tiles and further into sub-tiles, reducing the computational burden of processing the entire high-resolution image at once. The attention mechanism then selectively processes and weights the most relevant sub-tiles, maintaining diagnostic precision by focusing computational resources on critical regions rather than uniformly processing all high-resolution data.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent applies partial action by processing only a subset of sub-tiles in detail rather than all tiles uniformly. The attention mechanism identifies the most relevant sub-tiles that require detailed processing, while less relevant sub-tiles receive minimal or no detailed processing. This partial processing approach maintains diagnostic precision for critical regions while significantly reducing overall computational resource consumption.

Inventive Principle:
Principle #16Partial or excessive action

Data Source

PatentEP4645252A1Attention-based methods and systems for improving quality control of whole-slide image predictions
Publication Date: 2025.11.05 TEMPUS AI INC
  • EP4645252A1 patent drawingFigure 1
  • EP4645252A1 patent drawingFigure 2A
  • EP4645252A1 patent drawingFigure 2B

AI summary

A method involves receiving a whole-slide image, processing it with a machine learning model to generate a prediction, determining attention scores for image tiles, selecting a subset based on these scores, generating a pass/fail indication. A system includes processors and memory to perform these steps. A non-transitory computer-readable medium contains instructions for executing these processes.