Image Summarization for Endoscopy Using Deformation Analysis
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In medical imaging, particularly with endoscopic images, existing image summarization methods are inefficient as they often delete images containing non-lesion areas, leading to missed observations and increased burden on users due to reliance on lesion detection alone, without considering the importance of other high-observation-value areas like mucous membranes or areas with good visibility.
Innovation Solution
An image processing device and method that acquires an image sequence, detects observation target areas, selects reference and determination target images, calculates deformation information, and determines whether to delete images based on the observation target areas and deformation information to ensure critical areas are preserved in the summary sequence.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If image summarization is performed based only on lesion detection, then the number of images is reduced, but important non-lesion areas may be missed
Solution Approach 1:
The patent applies local quality by differentiating between different types of areas in the image (lesion areas, bubble areas, residue areas, dark areas, halation areas) and assigning different importance levels to each. The system selectively preserves images based on the presence of high-observation-value areas while allowing deletion of images containing only low-value areas, thus achieving both compression and reliability.
Solution Approach 2:
The patent changes the parameter used for image selection from binary lesion detection to a multi-parameter evaluation system that considers lesion area, bubble area, residue area, dark area, and halation area. This parameter transformation enables more nuanced decision-making about which images to preserve.
2Reliability
If all images in the sequence are checked, then observation completeness is maintained, but user time and burden increase significantly
Solution Approach 1:
The patent extracts and identifies high-observation-value areas (lesion, non-bubble, non-residue, non-dark, non-halation areas) from each image and uses these extracted features as the basis for summarization decisions. This extraction approach allows the system to focus only on critical areas rather than requiring manual review of entire images.
Solution Approach 2:
The patent introduces an intermediary processing step that automatically evaluates images based on multiple area parameters and generates a summary sequence. This intermediary system acts as a filter between the full image sequence and the user, providing reliable observation coverage while eliminating the need for users to review all images.
3Manufacturing precision
If images with bubbles or residues are deleted, then image quality is improved, but areas that cannot be observed are increased
Solution Approach 1:
The patent inverts the conventional approach by not deleting images with bubbles or residues, but rather using the presence of these areas as a factor to determine image preservation. Images containing only bubble or residue areas (without significant lesion or other high-value areas) are candidates for deletion, while images with both bubbles/residues and important features are preserved.
4Illumination intensity
If images with dark areas or halation areas are deleted, then visibility is improved, but important areas may be lost
Solution Approach 1:
The patent applies local quality by specifically identifying and evaluating dark areas and halation areas separately from other image features. Rather than uniformly deleting images with poor visibility, the system assesses whether important features (lesions, etc.) are present in or near these areas and makes preservation decisions accordingly.
Data Source
AI summary
An image processing device includes: an image sequence acquisition section that acquires an image sequence that includes a plurality of constituent images; and a processing section that performs an image summarization process that deletes some of the plurality of constituent images included in the image sequence to generate a summary image sequence, the processing section detecting an observation target area from each of the plurality of constituent images, selecting a reference image and a determination target image from the plurality of constituent images, calculating deformation information about a deformation estimation target area included in the reference image and the deformation estimation target area included in the determination target image, and determining whether or not the determination target image can be deleted based on the observation target area included in the reference image, the observation target area included in the determination target image, and the deformation information.