AI Image Labeling via Block Segmentation for Pathology
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current artificial intelligence-based image processing methods for lymph node metastasis in cancer diagnosis are inefficient and inaccurate due to the labor-intensive and time-consuming process of manual labeling of tens of thousands of pixels in pathological images, often resulting in conflicting region boundaries.
Innovation Solution
An AI-based image processing method that divides a slice image into multiple blocks, feeds them into a labeling model to extract pixel features, determine vertex positions of a polygonal region, and output label information, reducing the need for pixel-wise prediction and manual labeling.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If manual labeling of complete pathological images is performed, then label information can be obtained for training AI models, but the process becomes extremely time-consuming and labor-intensive
Solution Approach 1:
The patent divides a complete pathological image into multiple smaller image blocks, processes them in parallel, and combines results. This segmentation approach maintains labeling accuracy while dramatically reducing the time required, as multiple blocks can be processed simultaneously rather than requiring sequential pixel-wise analysis of the entire large-scale image.
Solution Approach 2:
The patent performs preliminary processing by dividing the image into blocks and identifying candidate regions before final label generation. This preliminary segmentation and block-based processing prepares the data structure in advance, enabling faster final labeling without sacrificing accuracy.
2Manufacturing precision
If pixel-wise manual labeling is performed on complete pathological images, then detailed label information is obtained, but the complexity and labor cost increase significantly
Solution Approach 1:
By segmenting the complete pathological image into multiple smaller blocks, the patent reduces the complexity of processing each individual block while maintaining overall labeling detail precision. The segmentation allows the system to handle complex labeling tasks in manageable units that can be processed more efficiently.
Solution Approach 2:
The patent uses a unified labeling model that can be applied consistently across all image blocks. This model copying approach ensures that the same labeling logic and precision standards are replicated across all blocks, maintaining detail precision without requiring separate complex processing for each region.
3Loss of information
If manual labeling of complete pathological images is performed, then comprehensive label information is obtained, but labor costs and processing time increase significantly
Solution Approach 1:
The patent segments the complete pathological image into multiple blocks that are processed in parallel. This segmentation maintains information completeness by ensuring all blocks are processed and combined, while dramatically improving productivity through parallel processing of multiple blocks simultaneously rather than sequential pixel-wise analysis.
Solution Approach 2:
The patent merges the results from multiple processed image blocks to generate the final comprehensive label information for the complete pathological image. This merging process ensures that all label information is preserved and integrated, maintaining completeness while achieving high productivity through the parallel processing architecture.
4Measurement precision
If traditional labeling methods are used, then label information can be generated, but the labeled regions often have conflicting boundaries and pores
Solution Approach 1:
By processing the image in segmented blocks and then merging results, the patent ensures that boundary decisions are made at multiple scales. This multi-scale segmentation approach resolves boundary conflicts by allowing local optimizations in each block while maintaining global consistency through the merging process, eliminating pores and ensuring continuous region boundaries.
Data Source
AI summary
This application discloses an artificial intelligence-based image processing method, apparatus, device, and storage medium, and relates to the field of computer technology. The method includes: obtaining a slice image; dividing the slice image to obtain a plurality of image blocks; feeding the plurality of image blocks into a labeling model, extracting, by the labeling model, a pixel feature of the slice image based on the plurality of image blocks, determining a plurality of vertex positions of a polygonal region in the slice image based on the pixel feature, concatenating the plurality of vertex positions, and outputting label information of the slice image, the polygonal region being a region in which a target pathological tissue of interest is located.