Slide Grid Classification for Faster Whole Slide Imaging
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Whole slide imaging (WSI) requires efficient implementation of grid rejection and extension techniques to reduce scan time, avoid scanning non-interesting regions, and ensure accurate localization and classification of tissue segments.
Innovation Solution
An apparatus and method using machine learning to identify and classify areas of interest on a slide by capturing macro and high-magnification images, employing a grid extension model to conditionally extend borders based on content detection, and utilizing a classifier model to distinguish between accepted and rejected grids.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If high magnification scanning is performed on all areas of the slide, then comprehensive analysis accuracy is improved, but scan time increases significantly
Solution Approach 1:
The slide is divided into multiple grids, and further segmented into regions of interest (ROI) and non-ROI areas based on macro image analysis. This segmentation allows high magnification scanning to be applied selectively only to relevant regions, maintaining comprehensive analysis accuracy while significantly reducing overall scan time by excluding non-interesting areas from detailed examination.
Solution Approach 2:
A macro image scan is performed first to identify regions of interest before conducting high magnification scanning. This preliminary action enables the system to pre-determine which areas require detailed examination, avoiding unnecessary high magnification scans of non-interesting regions and thereby reducing total scan time while preserving accuracy for relevant areas.
2Productivity
If grid rejection techniques are implemented to skip non-interesting regions, then scan time is reduced, but risk of omitting faint peripheral tissues increases
Solution Approach 1:
The system applies different scanning qualities to different regions: high magnification scanning is applied to regions of interest while lower magnification or no scanning is applied to non-ROI areas. This local quality differentiation maintains high detection reliability for relevant tissues while improving scan efficiency by skipping unnecessary areas. The macro image serves as a guide to determine which regions receive high-quality scanning.
Solution Approach 2:
The system uses feedback from macro image analysis to dynamically adjust scanning decisions. The macro image provides information about tissue distribution and potential regions of interest, which feeds into the grid classification process. This feedback mechanism ensures that faint peripheral tissues are not omitted by providing continuous information about where detailed scanning is most likely to yield meaningful results.
3Loss of time
If macro images are used to guide high magnification scanning, then scan time is reduced, but accurate localization of tissue segments becomes more difficult
Solution Approach 1:
The system introduces an intermediary classification process that bridges the macro image and high magnification scanning. The grid extension model and classifier model act as intermediaries that analyze the macro image, identify regions of interest, and generate a map guiding high magnification scanning. This intermediary layer preserves localization accuracy by systematically processing macro image information to accurately identify and locate tissue segments before detailed scanning begins.
Data Source
AI summary
An apparatus and method for detecting content of interest on a slide using machine learning. The apparatus includes at least a processor and a memory communicatively connected to the at least a processor. The memory instructs the processor to receive a first image, comprising a macro image, identify areas of interest associated with the grids of the first image, receive a second image comprising a high magnification image associated with the areas of interest of the first image, classify, using at least a probed point, the grids of the first image, wherein classifying the grids of the first image includes classifying the grids into accepted grids of the grids and rejected grids of the grids, scan, using the image capturing device, the accepted grids to generate an output image, and display, using a display device, the output image.