Medical Image Augmentation With Consistent Examination Results
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing medical image processing systems face challenges in improving prediction accuracy due to limited learning data, and combining medical images and examination results in multimodal learning requires maintaining consistency in generated examination results.
Innovation Solution
An image processing apparatus and method that generates new medical images and examination results by performing non-linear registration and transformation of existing images, using transformation coefficients based on normal distributions to maintain consistency and improve learning data variety.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Quantity of substance
If data augmentation methods (parallel displacement, rotation, enlargement/reduction, inversion, cropping, noise addition) are applied to generate new images, then the number of learning images is increased, but the variation of learning images is not sufficiently increased since only images similar to the target image are generated
Solution Approach 1:
The patent combines multiple data augmentation techniques including parallel displacement, rotation, enlargement/reduction, inversion, cropping, noise addition, and blurring into a comprehensive processing pipeline. By merging these different transformation methods, the system generates diverse learning images with sufficient variation while maintaining consistency with the target image characteristics
Solution Approach 2:
The patent applies parameter changes by systematically varying transformation parameters such as displacement amounts, rotation angles, enlargement/reduction ratios, noise levels, and blur degrees. These parameter variations enable generation of diverse learning images with different characteristics while maintaining controlled variation suitable for medical image analysis
2Measurement precision
If multimodal learning is implemented by combining medical images and examination results, then prediction accuracy of disease opinion is improved, but consistency must be maintained between new examination results and original examination results when generating new images
Solution Approach 1:
The patent implements feedback mechanisms by comparing generated new examination results with original examination results to ensure consistency. The system uses the original examination results as reference feedback to validate and adjust the generated data, maintaining reliability while enabling multimodal learning improvements
Solution Approach 2:
The patent performs preliminary processing and validation of examination results before final generation. By preparing and verifying consistency conditions in advance, the system ensures that new examination results maintain reliability with original data while enabling improved prediction accuracy through multimodal learning
3Measurement precision
If the number of learning images is increased to improve prediction accuracy, then more learning time and computational resources are required
Solution Approach 1:
The patent creates synthetic copies of medical images through various transformation operations (parallel displacement, rotation, enlargement/reduction, inversion, cropping, noise addition, blurring). These copied and transformed images serve as additional learning data without requiring extra acquisition time, thereby improving prediction accuracy while avoiding proportional increases in learning time
Data Source
AI summary
An image processing apparatus includes: a processor; and a memory connected to or built in the processor. The processor is configured to generate, as learning data used for training a machine learning model for medical images and examination results of medical examinations, new medical images from a first medical image and a second medical image among a plurality of the medical images according to a generation condition, and generate new examination results by performing calculation based on the generation condition on a first examination result of the medical examination corresponding to the first medical image and a second examination result of the medical examination corresponding to the second medical image.