CNN Segmentation Mask for Digital Pathology Areas of Interest
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current digital pathology methods for detecting tumors and toxicological changes in tissue samples are inefficient and prone to errors, requiring extensive manual review by pathologists and lacking effective automated tools to accurately identify areas of clinical interest.
Innovation Solution
A convolutional neural network (CNN) is trained on histological images and pathologist interaction data to classify pixels into relevance classes, generating a segmentation mask that highlights areas of interest, thereby guiding pathologists to relevant regions and improving analysis efficiency.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If manual review by pathologists is used to detect tumors and toxicological changes, then diagnostic accuracy can be maintained, but analysis efficiency and productivity are reduced
Solution Approach 1:
The patent segments the large histological image into multiple smaller patches, which are then processed individually by the CNN. This segmentation allows the system to efficiently analyze large images while maintaining diagnostic accuracy through focused examination of relevant regions.
Solution Approach 2:
The patent introduces a CNN-based automated analysis system as an intermediary tool that assists pathologists. The CNN generates probability maps and identifies areas of interest, which then guide the pathologist's review process, combining automated efficiency with human diagnostic expertise.
2Productivity
If automated CNN methods are used to identify areas of interest, then analysis speed and productivity improve, but reliability and accuracy may deteriorate due to false positives or missed areas
Solution Approach 1:
The CNN performs preliminary analysis by generating probability maps and identifying potential areas of interest before the pathologist begins detailed review. This preliminary action filters and prioritizes regions, allowing the pathologist to focus on the most suspicious areas while maintaining comprehensive coverage.
Solution Approach 2:
The system provides visual feedback through probability maps and area of interest annotations that guide the pathologist's review process. The pathologist can verify, reject, or refine the CNN's findings, creating a feedback loop that improves both speed and accuracy.
3Reliability
If pathologists review entire histological images comprehensively, then diagnostic reliability is maintained, but time consumption and loss of time increase
Solution Approach 1:
The patent applies local quality by focusing detailed analysis on specific regions of interest rather than uniformly processing the entire image. The CNN identifies and highlights areas with abnormal features, allowing the pathologist to allocate more time and attention to these critical regions while reducing review time for normal areas.
Solution Approach 2:
The CNN performs preliminary identification of suspicious areas, creating a prioritized list of regions that require detailed pathologist review. This preliminary action eliminates the need for pathologists to manually scan entire images, significantly reducing time loss while maintaining comprehensive review quality through targeted examination.
4Measurement precision
If multiple magnification levels are used for detailed examination, then measurement precision and detection accuracy improve, but device complexity and operational complexity increase
Solution Approach 1:
The CNN acts as an intermediary that processes the entire image at one magnification level and identifies regions requiring detailed examination. This eliminates the need for pathologists to manually switch between multiple magnification levels, reducing operational complexity while maintaining the ability to examine details at higher magnification when needed.
Solution Approach 2:
The system performs preliminary analysis at low magnification to identify areas of interest, then automatically guides the pathologist to these regions for detailed examination at higher magnification. This preliminary action reduces the number of magnification switches needed compared to traditional comprehensive review methods.
Data Source
AI summary
A CNN is applied to a histological image to identify areas of interest. The CNN classifies pixels according to relevance classes including one or more classes indicating levels of interest and at least one class indicating lack of interest. The CNN is trained on a training data set including data which has recorded how pathologists have interacted with visualizations of histological images. In the trained CNN, the in-terest-based pixel classification is used to generate a segmentation mask that defines areas of interest. The mask can be used to indicate where in an image clinically relevant features may be located. Further, it can be used to guide variable data compression of the histological image. Moreover, it can be used to control loading of image data in either a client-server model or within a memory cache policy. Furthermore, a histological image of a tissue sample of a tissue type that has been treated with a test compound is image processed in order to detect areas where toxic reactions to the test compound may have occurred. An autoencoder is trained with a training data set comprising histological images of tissue samples which are of the given tissue type, but which have not been treated with the test compound. The trained autoencoder is applied to detect tissue areas by their deviation from the normal variation seen in that tissue type as learnt by the training process, and so build up a toxicity map of the image. The toxicity map can then be used to direct a toxicological pathologist to examine the areas identified by the autoencoder as lying outside the normal range of heterogeneity for the tissue type. This makes the pathologist's review quicker and more reliable. The toxicity map can also be overlayed with the segmentation mask indicating areas of interest. When an area of interest and an area identified as lying outside the normal range of heterogeneity for the tissue type, and increased confidence score is applied to the overlapping area.