Medical Image Processing with Reference Images and Unpaired Training
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing medical image diagnostic systems face challenges in optimizing image quality due to the rarity of paired input and target images, lack of transparency in machine learning model processes, and the limited application of non-reference-type indices like RISQUE for medical images.
Innovation Solution
A medical image processing apparatus and method that uses a trained machine learning model, such as CycleGAN, to generate a reference image from an input image, adjusting image quality based on a comparison with a reference image, ensuring transparency and explainability through unpaired training data and conventional image filters.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a machine learning model is used to generate images with reduced noise, then image quality is improved, but transparency and explainability of the generation process deteriorate
Solution Approach 1:
The patent introduces a reference image as an intermediary element that mediates between the input image and the machine learning model's generation process. The reference image serves as a transparent benchmark that allows clinicians to understand and verify the transformation, making the black-box ML process interpretable through visual comparison and quality metric analysis
2Measurement precision
If paired input and target images are used for training, then optimization of image quality is improved, but availability of training data deteriorates
Solution Approach 1:
The patent inverts the traditional paired training approach by using unpaired images for training the machine learning model. Instead of requiring corresponding input-target image pairs, the system trains on independent datasets and uses a reference image for quality assessment, thereby expanding the available training data while maintaining optimization capability
Solution Approach 2:
The reference image acts as an intermediary that enables quality optimization without requiring paired training data. It serves as a standalone benchmark that the generated images are compared against, allowing the system to learn from unpaired data while maintaining quality standards through the mediating reference
Data Source
AI summary
A medical image processing apparatus of an embodiment includes processing circuitry. The processing circuitry acquires an input image to be processed. The processing circuitry generates a reference image from the input image using a trained model. The processing circuitry adjusts the image quality of the input image. The processing circuitry compares the image-quality-adjusted image, which is the input image with the adjusted image quality, with the reference image and calculates an image quality difference between the images. The trained model is a machine learning model trained on the basis of a training data set including two unpaired images. The processing circuitry readjusts the image quality of the input image on the basis of the image quality difference.