Bioimage Converter Using Classifier Feedback to Remove Metal Artifacts
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing bioimage acquisition technologies fail to precisely obtain a positive bioimage with few defects from a negative bioimage containing metal artifacts, as they require supervised learning with paired defective and non-defective images, which are difficult to obtain.
Innovation Solution
A bioimage acquiring device comprising a classifier to distinguish between negative and positive bioimages, a conversion unit to update the converter based on the classification results, and a learning process that reduces feature difference information, allowing for the acquisition of a positive bioimage with few defects without using teacher data.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If supervised learning is used to convert negative bioimages to positive bioimages, then conversion capability is achieved, but the requirement for paired teacher data (defective and non-defective image pairs) becomes a bottleneck
Solution Approach 1:
The system performs self-supervised learning by automatically generating training data from available negative bioimages without requiring external paired positive bioimages. The converter learns to reduce metal artifacts through self-assessment and iterative optimization, eliminating the need for difficult-to-obtain teacher data while maintaining high conversion precision
Solution Approach 2:
A classifier is introduced as an intermediary component that evaluates the quality of converted bioimages and provides feedback signals to the converter. This mediator enables the system to learn from unpaired data by using the classifier's assessments as surrogate supervision, replacing the need for actual paired training images
2Object-generated harmful factors
If conventional MAR algorithms or deep learning techniques are used, then some artifact reduction is achieved, but precise acquisition of positive bioimages with few defects cannot be obtained
Solution Approach 1:
The system implements a feedback loop where the classifier evaluates converted bioimages and provides determination results back to the converter for iterative optimization. This feedback mechanism enables continuous refinement of the conversion process, progressively reducing metal artifacts and improving the precision of the acquired positive bioimages until high-quality defect-free images are obtained
Solution Approach 2:
The converter is designed as a dynamic, learnable model that adapts its parameters through iterative optimization based on classifier feedback. Unlike static conventional MAR algorithms, the converter dynamically adjusts its transformation strategy to precisely eliminate artifacts while preserving anatomical features, achieving superior defect reduction precision
Data Source
AI summary
A bioimage acquiring device includes a first conversion unit that performs a first conversion process wherein a negative bioimage is converted using a first converter so as to acquire a converted bioimage which is the result of the conversion, and a classifying unit that performs a first classifying process wherein a determination is made as to whether the converted bioimage is a positive bioimage or a negative bioimage, using a classifier for determining whether an image is a negative bioimage or a positive bioimage, wherein the first conversion unit performs a learning process using the determination result obtained by the classifying unit and the converted bioimage, performs an update process for updating the first converter, and receives a new negative bioimage, and the first conversion unit converts the new negative bioimage using the updated first converter to acquire a converted bioimage which is the result of the conversion.