Neural Network Region Extraction with Interactive Confidence Adjustment
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current image processing systems for assessing pressure ulcers face challenges in accurately extracting and adjusting the affected area region, particularly when using machine learning methods with insufficient training data or when there are no reference structures available, leading to inconsistencies between user determinations and automated extractions.
Innovation Solution
An image processing apparatus that includes an acquisition unit for image data, a calculation unit to determine confidence values for each pixel using a trained neural network, a modification unit to adjust reference values, and a display unit to visually assist users in extracting and adjusting the region of interest, allowing for interactive refinement of the affected area extraction.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If a neural network model is used to automatically extract the affected area region, then the extraction speed and consistency are improved, but the accuracy and adaptability to different assessment goals deteriorate when training data is insufficient or inaccurate
Solution Approach 1:
The patent divides the affected area extraction into two stages: an automatic extraction stage using a neural network model to provide an initial extraction, and a manual adjustment stage where the user refines the extraction by selecting specific pixels. This segmentation allows the system to leverage automated processing for speed while reserving manual intervention for accuracy, effectively resolving the contradiction between extraction speed and accuracy.
Solution Approach 2:
The patent implements a feedback mechanism where the user's manual pixel selection and region adjustment provide corrective feedback to the system. The user can visually review the automatically extracted region and make corrections by selecting pixels that should be included or excluded, thereby improving the accuracy of the extraction while maintaining the efficiency of the automated initial extraction.
2Loss of time
If the affected area region is extracted using a fixed classification model, then the processing time is reduced, but the adaptability to different assessment purposes deteriorates
Solution Approach 1:
The patent introduces dynamic adaptability by allowing the user to adjust the extracted region based on specific assessment goals. Instead of a fixed extraction, the system enables users to modify the region boundaries and pixel selections according to their particular assessment requirements, such as focusing on the center of the affected area or extending to the periphery, thereby achieving both speed and adaptability.
Solution Approach 2:
The patent applies local quality by allowing different regions of the affected area to be treated differently based on assessment needs. Users can selectively adjust specific areas of the extraction, such as refining the boundaries in certain regions while maintaining others, enabling the system to adapt to various assessment goals without requiring complete re-extraction.
3Measurement precision
If manual measurement methods are used to ensure accuracy, then the measurement precision is improved, but the time consumption and operational complexity increase
Solution Approach 1:
The patent applies preliminary action by having the neural network model perform the bulk of the extraction work automatically before the user intervenes. The model provides an initial extraction that covers the majority of the affected area, and the user only needs to make minor adjustments by selecting specific pixels, significantly reducing the time compared to complete manual measurement while maintaining high accuracy.
4Ease of operation
If the extraction region is defined by color gamut alone, then the ease of operation is improved, but the reliability of the extraction deteriorates when regions with similar colors are mistakenly included
Solution Approach 1:
The patent introduces an intermediary mechanism where the user's pixel selection serves as a mediator between the automatic extraction and the final result. The system uses the user's interactive input to resolve ambiguities in color-based extraction, allowing the user to confirm or correct the inclusion of regions with similar colors, thereby improving reliability while maintaining ease of operation through simple pixel selection.
Data Source
AI summary
An image processing apparatus includes an acquisition unit configured to acquire image data, a calculation unit configured to calculate, for each unit region of the image data, a confidence that the unit region is an extraction subject, the confidence as the extraction subject being calculated for each unit region of the image data by inputting the image data into a trained model of a neural network that has been trained using images of an existing extraction subject or a region of interest as training data, a modification unit configured to modify a reference value of the confidence, which is used to extract a region of interest, an extraction unit configured to extract the region of interest on the basis of the calculated confidence of each unit region and the modified reference value, and a display unit configured to display the extracted region of interest.