Weakly Supervised Medical Image Abnormality Detection via Down-Sampling Classification
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current image processing methods require significant manpower and resources for abnormality classification, especially in fields like medicine, where accurate annotation of lesion locations is challenging, leading to inefficiencies and inaccuracies.
Innovation Solution
The proposed method employs a weakly supervised learning approach using a computer vision model that performs down-sampling abnormality classification and up-sampling positioning processing to efficiently detect abnormalities in images, reducing the need for manual annotation and improving accuracy by using a convolutional neural network-based image detection model.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If dedicated technicians manually analyze images to determine abnormality categories and locations, then measurement precision is improved, but productivity deteriorates due to large manpower consumption and low inspection efficiency
Solution Approach 1:
The patent replaces the mechanical manual analysis system with an automated computer vision system. The image processing apparatus uses down-sampling abnormality classification processing and up-sampling abnormality positioning processing to automatically detect and locate abnormalities, substituting dedicated technicians with an automated computational system that maintains high accuracy while dramatically improving inspection efficiency
Solution Approach 2:
The patent segments the image processing task into two distinct stages: abnormality classification (down-sampling) and abnormality positioning (up-sampling). This segmentation allows each stage to be optimized independently, with the classification stage focusing on accurate abnormality identification and the positioning stage focusing on precise location determination, thereby resolving the contradiction between accuracy and efficiency
2Reliability
If dedicated technicians manually analyze images for abnormality classification, then reliability is improved through expert judgment, but loss of time increases due to manual processing requirements
Solution Approach 1:
The patent performs preliminary abnormality classification through down-sampling processing before conducting detailed abnormality positioning. This preliminary action efficiently identifies potential abnormalities and their categories, allowing the subsequent up-sampling positioning stage to focus only on regions of interest, thereby reducing overall processing time while maintaining reliable classification through the initial classification stage
3Productivity
If computer vision technology is applied to image detection, then productivity is improved through automated processing, but measurement precision may deteriorate compared to expert technician analysis
Solution Approach 1:
The patent changes the processing parameters by implementing a two-stage approach with different sampling rates. The down-sampling stage processes images at lower resolution for efficient classification, while the up-sampling stage restores detail for precise positioning. This parameter change allows the computer vision system to achieve both high productivity through automated processing and high measurement precision through multi-resolution analysis
Data Source
AI summary
Embodiments of this application disclose an image processing method performed by a computer device, and a computer-readable storage medium. The method includes: obtaining a to-be-detected image, and performing down-sampling abnormality classification processing on the to-be-detected image, to obtain a predicted abnormality category label and a target feature image; performing preliminary abnormality positioning processing based on the predicted abnormality category label and the target feature image, to obtain an initial positioning image corresponding to the to-be-detected image; performing up-sampling abnormality positioning processing on the initial positioning image, to obtain a target positioning image corresponding to the to-be-detected image; and outputting the predicted abnormality category label and the target positioning image, the initial positioning image and the target positioning image being configured for reflecting attribute information of a target region associated with the predicted abnormality category label within the to-be-detected image.