Image Processing Apparatus for Medical Correct Answer Image Generation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Creating accurate correct answer images for machine learning in the medical field is labor-intensive, especially when dealing with large frame groups, as existing methods require manual masking of lesion regions in each frame.
Innovation Solution
An image processing apparatus that selects reference frames based on image quality, acquires reference correct answer frames, and creates complementary correct answer frames through morphing processes, reducing the need for manual intervention and efficiently generating correct answer image groups for machine learning.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Manufacturing precision
If manual masking process is performed for each frame in a large frame group, then accurate correct answer images can be created, but a lot of effort and time are necessary
Solution Approach 1:
The patent segments the frame group into a reference frame and multiple target frames. The reference frame is manually processed to create a reference correct answer frame, while target frames are automatically processed using the reference correct answer frame as a template. This segmentation reduces manual effort from processing every frame to only processing the reference frame.
Solution Approach 2:
The patent performs preliminary manual masking on the reference frame to create the reference correct answer frame before processing target frames. This preliminary action establishes a template that guides the automatic creation of correct answer frames for all other frames, eliminating the need for repeated manual intervention.
Solution Approach 3:
The patent copies the reference correct answer frame to create correct answer frames for target frames. By using the reference correct answer frame as a template and applying it to target frames through image processing techniques, the system automatically generates accurate correct answer images without manual masking for each frame.
2Measurement precision
If manual masking is performed for every frame, then correct answer images with accurate lesion regions can be obtained, but the work process becomes extremely time-consuming
Solution Approach 1:
The patent divides the frame group into one reference frame and multiple target frames. Only the reference frame requires manual lesion region designation, while target frames use automatic designation based on the reference frame. This segmentation reduces time investment from proportional to total frame count to a fixed small amount.
Solution Approach 2:
The patent changes the parameter of lesion region designation from manual interaction for each frame to automatic computation using image processing algorithms. By transforming the designation process from a manual parameter-setting task to an automated algorithmic process, time consumption is dramatically reduced while maintaining accuracy.
3Reliability
If correct answer images are created manually for machine learning training, then high quality training data can be obtained, but the complexity of the work process increases
Solution Approach 1:
The patent segments the image processing workflow into a simple reference frame processing stage and an automated target frame processing stage. This segmentation simplifies the overall workflow complexity by separating manual intervention (needed only once) from automated processing (applied to all frames).
Solution Approach 2:
The patent creates a universal reference correct answer frame that serves as a template for all target frames. This single reference correct answer frame performs multiple functions: it defines the lesion region for the reference frame and serves as the basis for generating correct answer frames for all target frames through automatic processing.
Data Source
AI summary
An image processing apparatus having a processor configured to: select a reference frame from a frame group including a plurality of images; acquire a reference correct answer frame representing a region of interest in the selected reference frame; generate a complementary correct answer frame corresponding to a frame other than the reference frame included in the frame group based on at least one reference correct answer frame; and generate a correct answer image group for machine learning from the reference correct answer frame and the complementary correct answer frame.