Whole Slide Image Feature Enhancement for Pathology
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In digital pathology, there is a mismatch between features visible at different magnification levels, making it difficult for pathologists to identify and analyze specific features, such as hotspots, due to the subjective nature of diagnostic tasks and the limitations of current image processing methods.
Innovation Solution
The method programmatically enhances features of interest in digital pathology images by extracting and modifying high-magnification features to make them visible in low-magnification views, and vice versa, using techniques like color deconvolution, blurring, and subsampling, to create visually enhanced images that aid in diagnostic tasks.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If standard subsampling is used to generate low magnification views, then processing speed is improved, but diagnostic accuracy deteriorates due to loss of high-magnification features
Solution Approach 1:
The image processing is segmented into multiple magnification levels (low, medium, high), with each level processed and enhanced independently. High-magnification features are extracted, enhanced, and then integrated into low-magnification views, allowing diagnostic features to be preserved across different scales without reprocessing the entire high-resolution image.
Solution Approach 2:
Different regions of the image are processed with different quality levels based on their diagnostic importance. Regions containing high-magnification diagnostic features (such as hotspots with concentrated stained cells) are enhanced and preserved at higher quality, while other regions use standard subsampling, optimizing both speed and diagnostic accuracy locally.
2Measurement precision
If high magnification features are preserved in low magnification views, then diagnostic accuracy is improved, but data processing complexity increases
Solution Approach 1:
High-magnification features are identified and extracted in advance during the image acquisition phase. These features are pre-processed and stored separately, so they can be quickly integrated into low-magnification views without requiring complex real-time processing of the entire high-resolution image.
Solution Approach 2:
Instead of processing and displaying the entire high-resolution image at all magnification levels, the system creates simplified copies or representations of high-magnification features (such as extracted cell locations, hotspot regions, or feature descriptors) that can be overlaid on low-magnification views, reducing computational complexity while preserving diagnostic information.
3Ease of operation
If feature enhancement is applied to make high-magnification features visible in low-magnification views, then ease of operation is improved, but image processing time increases
Solution Approach 1:
Feature enhancement is applied selectively only to regions containing diagnostically important high-magnification features (such as hotspots or regions with concentrated stained cells), rather than processing the entire image. This partial application of enhancement maintains ease of operation for critical areas while minimizing overall processing time.
Data Source
Figure 1
Figure 2A~3B
Figure 4A~5C
AI summary
A method of processing digital pathology and cytology images for viewing, a computer program product and a system for evaluating Whole Slide Images, "WSI", are provided. An original digital pathology or cytology image of a tissue sample is obtained. Features of interest in the obtained image that are otherwise only visually apparent in a high magnification sub-slide view are programmatically visually enhanced to be visually apparent in a low magnification whole slide view, and/or features of interest are programmatically visually enhanced using data from a low magnification sub-slide view to generate a high magnification whole slide view. The visually enhanced features are electronically displayed.