Endoscopic Image Processing for Swallowing Phase Classification
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing video endoscopic examinations for swallowing function evaluation face challenges in accurately determining the progression of swallowing movements, leading to oversight and increased user burden due to the need for reviewing large volumes of images, with existing methods struggling to differentiate similar images or values across different stages of swallowing.
Innovation Solution
An image processing device that classifies each frame of the examination image into swallowing-in-progress or not-in-progress frames, integrates adjacent frames within a specific temporal threshold as swallowing blocks, and uses threshold values to determine the progression of swallowing motions, incorporating deep learning and machine learning for accurate classification.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If each frame is individually classified using deep learning to determine swallowing progress, then classification accuracy improves, but the complexity of processing increases and oversight may still occur due to similar images across different time zones
Solution Approach 1:
The patent merges multiple individual frame classifications into integrated swallowing blocks by combining adjacent frames that meet temporal criteria. This reduces the overall complexity by processing groups of frames as unified units rather than individually, while maintaining accurate swallowing determination through the integration of multiple classification results.
Solution Approach 2:
The patent performs preliminary classification of each frame using deep learning to identify potential swallowing-in-progress frames before integrating them into swallowing blocks. This preliminary action enables accurate identification of candidate frames that can then be efficiently grouped, reducing subsequent processing complexity while maintaining high determination accuracy.
2Reliability
If all acquired images are reviewed to prevent oversight during examination, then determination reliability improves, but user burden increases due to the large number of images requiring observation
Solution Approach 1:
The patent segments the large volume of acquired images into smaller, manageable swallowing blocks based on temporal intervals and classification results. This segmentation allows users to review only the relevant blocks containing swallowing events rather than all images, maintaining examination reliability while significantly reducing user burden.
Solution Approach 2:
The patent extracts and highlights only the swallowing blocks from the complete image sequence, separating the relevant information (swallowing events) from the irrelevant content (non-swallowing frames). This extraction enables users to focus on critical frames for accurate determination without being overwhelmed by the total number of images.
3Productivity
If frames with temporal intervals equal to or less than a threshold are integrated into swallowing blocks, then productivity improves by reducing review burden, but measurement precision may decrease due to potential inclusion of similar but incorrect frames
Solution Approach 1:
The patent dynamically adjusts the integration criteria for swallowing blocks by evaluating both temporal intervals and classification results. Frames are integrated into swallowing blocks only when they meet both the temporal threshold and the swallowing-in-progress classification, creating a dynamic filtering mechanism that maintains precision while improving productivity through efficient grouping.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
The present invention provides an image processing device and an operation method of an image processing device capable of preventing oversight during an examination or reducing a burden on a user in a video endoscopic examination of swallowing. An image acquisition unit acquires an examination image. A first frame classification unit classifies each frame of the examination image into any one of a swallowing-in-progress frame in which swallowing is in progress or a swallowing-not-in-progress frame in which the swallowing is not in progress. A swallowing block generation unit integrates, in a case in which a temporal interval between the swallowing-in-progress frames temporally adjacent to each other is equal to or less than a first threshold value, at least the adjacent swallowing-in-progress frames as a swallowing block.