Composite-Tiled Image Embeddings for Machine Learning Efficiency
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing predictive data analysis solutions face efficiency and reliability challenges when processing large input images, particularly in reducing the size of input feature data while minimizing accuracy drawbacks from partial processing.
Innovation Solution
The use of composite-tiled image embeddings generated by merging multiple tiled images created using various tiling mechanisms, which reduces the size of input feature data and processing operations for image processing machine learning models, especially for large images like biopsy images.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If multiple tiling mechanisms are used to generate tiled images, then computational efficiency is improved through reduced input size, but accuracy may deteriorate due to partial processing of large input images
Solution Approach 1:
The input image is divided into multiple image regions that are then processed through different tiling mechanisms to generate multiple tiled images. Each tiling mechanism (e.g., greedy thresholding, sorted selection, sorted row-wise selection) creates a different segmentation pattern, allowing the system to capture diverse regional characteristics while maintaining computational efficiency through the selective representation of image content.
2Loss of time
If the size of input feature data is reduced through tiling, then processing speed is improved, but information loss occurs from partial processing
Solution Approach 1:
Multiple tiled images generated by different tiling mechanisms are merged into a single composite-tiled image that preserves the essential information from each tiling approach. This merging process creates a comprehensive representation that maintains critical image features while keeping the data size manageable for efficient machine learning processing.
Solution Approach 2:
The composite-tiled image acts as a composite structure that integrates multiple tiled images with different characteristics. By combining the strengths of various tiling mechanisms (greedy thresholding for prominent regions, sorted selection for diverse coverage, sorted row-wise selection for systematic sampling), the composite structure achieves both compression efficiency and information preservation.
3Device complexity
If N-sized tiled region subsets are selected from R image regions, then device complexity is reduced through smaller processing units, but measurement precision may deteriorate
Solution Approach 1:
The system varies the parameter N (number of regions to select) across different tiling mechanisms while keeping the total number of image regions R constant. This parameter variation allows the composite-tiled image to capture image characteristics at multiple scales and resolutions, improving measurement precision without requiring the processing unit to handle all R regions simultaneously.
Data Source
AI summary
As described herein, various embodiments of the present invention disclose techniques that improve efficiency of performing image-based machine learning operations on large images while limiting accuracy drawbacks of partial processing of those large images by using composite-tiled image embeddings for composite-tiled images generated by merging multiple tiled images that are generated using multiple tiling mechanisms. For example, in some embodiments, given an input image that comprises R image regions, each tiled image comprises N selected image regions of the R image regions that are selected in accordance with a tiling mechanism (where N<R). In this way, given T tiling mechansisms, T tiled images are generated, and the T tiled images are merged to generate a composite-tiled image. Accordingly, by using T tiling mechansisms, various embodiments enable reducing the size of feature data provided to an image processing machine learning model by selecting non-holistic subsets of an input image.