Data Augmentation for Medical Imaging with Defective Region Masking
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In medical radiation imaging, data augmentation techniques that rotate images can introduce defective regions with little to no image information, leading to reduced accuracy in region extraction due to the inclusion of areas where image information is uniformly set to arbitrary values, such as zero, which can hinder learning and lower performance.
Innovation Solution
A machine learning apparatus and method that performs data augmentation by increasing the number of input images while ensuring that regions with defective image information are not included, using affine transform processing, extraction processing, and signal amount adjustment to enhance generalization performance without introducing invalid regions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Quantity of substance
If data augmentation is performed by rotating images to multiple angles, then the number of training data is increased, but regions with defective image information are introduced leading to reduced learning accuracy
Solution Approach 1:
The patent segments the image into a valid region (where image information exists) and a defective region (where image information is uniform/arbitrary). By identifying and separating these regions, the patent applies different handling strategies: the valid region is used for normal data augmentation, while the defective region is excluded or masked to prevent it from degrading learning accuracy.
Solution Approach 2:
The patent applies different quality standards to different parts of the image. Instead of treating the entire rotated image uniformly, it identifies regions with sufficient image information quality versus regions with defective uniform values. The learning process then focuses on regions with adequate quality, allowing data augmentation to increase training data quantity without compromising overall learning accuracy.
2Stability of the object's composition
If arbitrary values such as zero are substituted into regions with defective image information, then the image structure is maintained, but the learning process is hindered and accuracy is lowered
Solution Approach 1:
The patent extracts or removes the defective regions with uniform arbitrary values from the training process. Instead of allowing these regions to interfere with learning, the system identifies them and excludes them from regions used for training the neural network, thereby preventing them from hindering the learning process while maintaining image structure integrity.
Solution Approach 2:
The patent introduces an intermediary mechanism (region identification and filtering process) between the rotated image and the learning process. This intermediary step identifies defective regions and prevents them from being used in training, acting as a mediator that allows data augmentation to proceed while blocking the harmful effect of defective regions on learning accuracy.
Data Source
AI summary
A machine learning apparatus for extracting a region from an input image, comprises: an inference unit configured to output the region by inference processing for the input image; and an augmentation unit configured to, in learning when learning of the inference unit is performed based on training data, perform data augmentation by increasing the number of input images constituting the training data, wherein the augmentation unit performs the data augmentation such that a region where image information held by the input image is defective is not included.