Multi-Resolution Tumor Segmentation in Digital Pathology
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for tumor detection and segmentation in digital pathology images are inefficient in resource usage and risk losing diagnostically relevant data, particularly due to the challenges of background removal and varying tissue textures, subtle cell differences, and mixed metastases, leading to missed clinically significant results.
Innovation Solution
The method involves creating a lower-resolution Whole Slide Image (WSI) for background removal using adaptive thresholding, followed by patch extraction and processing at optimized resolutions with machine learning systems for heat map creation and segmentation, ensuring more accurate and efficient tumor detection and segmentation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a single high resolution level is used for multiple image processing tasks, then detection precision is maintained, but computational resource demand increases significantly
Solution Approach 1:
The patent segments the image processing tasks by resolution level. It divides the whole slide image into multiple resolution levels (low, medium, high) and assigns different processing tasks to appropriate levels. Background removal and initial detection are performed on lower resolution images, while detailed segmentation and classification are performed on higher resolution patches. This segmentation allows the system to maintain detection precision where needed while reducing computational resource demand for preliminary processing steps.
Solution Approach 2:
The patent introduces the resolution dimension as an additional processing dimension. Instead of processing all tasks at a single resolution level, it creates a multi-resolution processing pipeline where images are processed at different resolution levels appropriate to each task. This dimensional approach allows the system to optimize resource usage by performing coarse processing at lower resolutions and fine processing at higher resolutions, thereby reducing overall computational demand while maintaining precision.
2Productivity
If a fixed intensity threshold is used for background removal, then processing speed is improved, but diagnostically relevant data is lost
Solution Approach 1:
The patent applies local quality by using adaptive thresholding that adjusts the intensity threshold based on local image characteristics rather than using a single fixed threshold for the entire image. The threshold is calculated locally for different regions, allowing the system to preserve cells with varying intensity values that would otherwise be lost. This local adaptation maintains processing efficiency while preventing loss of diagnostically relevant data in different image regions.
Solution Approach 2:
The patent changes the threshold parameter from a fixed value to a dynamically calculated value based on local image statistics. Instead of applying a constant intensity threshold, the system calculates thresholds adaptively for different regions, changing the parameter to match local conditions. This allows the system to maintain high processing speed while preserving cells with meaningful information that have intensity values near the threshold boundary.
3Measurement precision
If AI/machine learning is applied to process entire high resolution images, then detection accuracy is improved, but processing time increases significantly
Solution Approach 1:
The patent performs preliminary actions by conducting background removal and initial candidate identification on lower resolution images before applying computationally intensive AI/machine learning methods. This preliminary processing filters out obvious background regions and identifies potential regions of interest, so that the expensive AI processing is only applied to relevant patches rather than the entire high resolution image. This significantly reduces processing time while maintaining detection accuracy for the final results.
Solution Approach 2:
The patent segments the processing workflow into multiple stages with different resolution requirements. The first stage performs coarse processing (background removal, candidate identification) on lower resolution images. The second stage applies AI/machine learning methods only to selected patches at higher resolution. This segmentation allows the system to achieve high detection accuracy through AI while minimizing processing time by limiting AI application to relevant regions only.
Data Source
AI summary
A method of tumor detection and segmentation accepts a first Whole Slide Image (WSI) having a first resolution; creates a corresponding second WSI having a second resolution lower than the first resolution; applies an adaptive thresholding technique to the second WSI to create a background removal mask background; applies the mask to the first WSI to provide a third WSI with extracted patches, characterized by a third resolution, greater than the second resolution and lower than the first resolution; uses a first machine learning system on the third WSI to create a heat map at the third resolution, indicating a subset of the patches likely to include one or more clusters of tumor cells; and uses a second machine learning system on the first WSI and the heat map to segment each patch in a corresponding output image at the first resolution, outlining one or more corresponding clusters.