Image Preprocessing via Segmentation and Tiling for Whole-Slide Classification
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Histopathological image analysis in oncology is hindered by the cost and labor-intensive process of obtaining localized annotations, limiting dataset size and availability, especially for new disease subtypes, prognosis estimation, and drug response prediction, necessitating a method for accurate classification without localized annotation.
Innovation Solution
A method involving a device that segments images into regions of interest and background using a convolutional neural network, tiles these regions, extracts feature vectors, and processes them to classify images, reducing computational resources and enabling classification of large images without local annotations.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If localized annotation masks are used as training data, then classification accuracy is improved, but the cost and time required to obtain annotations increases significantly
Solution Approach 1:
The patent extracts only the essential diagnostic information (global classification labels) from the annotation process, eliminating the need for time-consuming pixel-by-pixel segmentation masks. By using only whole-slide image level labels rather than detailed localized annotations, the method achieves accurate classification while dramatically reducing the time and cost of data preparation.
Solution Approach 2:
The patent uses attention mechanisms to create a computational copy of the pathologist's diagnostic process, where the model learns to automatically identify and focus on relevant regions without requiring manual annotation of those regions. This copying approach allows the system to achieve high accuracy using only simple global labels.
2Measurement precision
If whole-slide images are processed at full resolution, then classification accuracy is improved, but memory requirements exceed available computational resources
Solution Approach 1:
The patent segments the whole-slide image into multiple smaller patches or tiles that can be processed individually within available memory constraints. The attention mechanism then learns to selectively focus on and aggregate information from the most relevant patches, maintaining classification accuracy while reducing peak memory requirements to manageable levels.
Solution Approach 2:
The patent transforms the problem from processing the entire 2D image at once to processing a collection of smaller 2D patches in a 3D tensor space (patches × features × channels). This dimensional transformation allows the model to handle large images by distributing computation across multiple smaller units that fit within memory constraints.
3Productivity
If traditional image processing methods are used, then computational resources are consumed, but the ability to handle large whole-slide images is limited
Solution Approach 1:
The patent applies local quality by enabling the model to process different regions of the whole-slide image with different levels of detail and computational resources. The attention mechanism identifies regions of interest that require detailed processing while allowing less important regions to be processed more coarsely, optimizing the balance between processing capability and resource consumption.
Data Source
AI summary
A method and apparatus of a device that classifies an image is described. In an exemplary embodiment, the device segments the image into a region of interest that includes information useful for classification and a background region by applying a first convolutional neural network. In addition, the device tiles the region of interest into a set of tiles. For each tile, the device extracts a feature vector of that tile by applying a second convolutional neural network, where the features of the feature vectors represent local descriptors of the tile. Furthermore, the device processes the extracted feature vectors of the set of tiles to classify the image.