Patch-Based Tissue Pathology Reading for Fewer False Negatives
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The increasing workload and shortage of pathologists, coupled with the limitations of artificial intelligence models, lead to human errors, particularly false negatives, in histopathologic diagnosis, especially in gastrointestinal pathology, where inter-observer variability and poor slide quality affect the accuracy of AI models.
Innovation Solution
A histopathologic reading support device and method that trains an AI model by dividing slide tissue images into patches, inferring classifications for each patch, and integrating these results to improve the accuracy of slide tissue image classification, using deep learning models like CNN to handle common artifacts and variability.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If high-resolution (×40 or higher) scanning is used to minimize the difference from traditional under-a-microscope reading, then diagnostic accuracy is improved, but significant costs are incurred in building and operating information computing infrastructure to store and process high-resolution slide tissue images
Solution Approach 1:
The patent divides high-resolution slide tissue images into multiple patches, allowing the AI model to process and analyze diagnostic information in smaller, manageable units. This segmentation enables accurate diagnosis while reducing the computational burden of processing entire high-resolution images at once.
Solution Approach 2:
The patent applies AI-based patch classification to only the necessary portions of slide images for diagnostic purposes, rather than requiring full high-resolution image processing for all cases. This partial action approach maintains diagnostic accuracy for critical areas while reducing overall computational resource consumption.
2Reliability
If AI models are used to assist in quality control and reduce false negatives, then diagnostic reliability is improved, but the complexity of the system increases due to integration of AI models with existing pathology workflows
Solution Approach 1:
The patent introduces an AI-based patch classification system as an intermediary layer between slide scanning and pathologist review. This intermediary automatically classifies patches and generates preliminary diagnostic information, reducing false negatives while maintaining a relatively simple integration with existing pathology workflows.
Solution Approach 2:
The AI model performs self-learning and automatic classification of tissue patches without requiring manual intervention for each case. The system automatically processes slides, identifies diagnostic features, and provides recommendations, reducing the need for complex human-AI coordination while improving reliability.
3Productivity
If AI classification models are applied as a screening tool in the pre-reading stage, then productivity is improved by prioritizing positive cases, but the accuracy may be affected by inter-observer variability and poor slide quality
Solution Approach 1:
The patent segments slide images into patches and applies AI classification to each patch independently. This approach allows the system to handle variability in slide quality by focusing on local features rather than requiring perfect overall image quality, thereby maintaining both productivity and accuracy.
Solution Approach 2:
The patent adjusts classification parameters and thresholds based on the specific characteristics of each patch and slide quality metrics. By dynamically changing parameters according to local conditions, the system maintains high screening efficiency while adapting to variations in slide quality and reducing the impact of inter-observer variability.
Data Source
AI summary
The present specification provides a tissue pathology reading support device and a method therefor, which, with respect to an artificial intelligence model, divide, into a plurality of patches, a slide tissue image for learning that indicates a lesion site if a lesion is present, infer the classification of each of the plurality of patches so as to learn patch classification results, integrate the plurality of patches and the patch classification results so as to infer the classification of the reconstructed slide tissue image, thereby learning slide tissue image classification results, use the trained artificial intelligence model so as to infer the classification of each of the plurality of patches, thereby generating patch classification results, and integrate the plurality of patches and the patch classification results so as to infer the classification of the reconstructed slide tissue image, thereby generating slide tissue image classification results.