Learning Device Noise Reduction Training Data Synthesis
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing noise reduction technologies for medical images struggle when image data with less noise than normal captured images is unavailable, and when the amount of noise in learning data is unknown.
Innovation Solution
A learning device that uses a trainable noise reducer to remove noise from first learning data, adds predetermined noise to the reduced data, and trains the noise reducer to align the distribution of the modified data with that of second learning data.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Manufacturing precision
If image data with small amount of noise is used for training the noise reducer, then the noise reduction performance is improved, but the acquisition difficulty increases because low-noise image data cannot be obtained in medical imaging without increasing exposure rate
Solution Approach 1:
The patent creates synthetic low-noise image data by copying and modifying existing noisy medical images. Specifically, it generates synthetic noisy images with controlled noise levels by adding noise to denoised images, thereby creating training data that mimics the target low-noise condition without requiring actual low-noise acquisitions. This resolves the contradiction by substituting difficult-to-obtain real low-noise data with synthesized copies that have similar statistical properties.
Solution Approach 2:
The patent introduces an intermediary approach by using synthetic images as a bridge between available noisy medical images and the desired low-noise training data. The synthetic images serve as mediators that transfer the statistical characteristics of low-noise images to the training process, enabling the noise reducer to learn effective denoising without direct access to actual low-noise medical images.
2Ease of manufacture
If only image data with the same noise as normal captured image is used for training, then the acquisition ease is improved, but the training effectiveness deteriorates because the amount of noise in the learning data must be known and controlled
Solution Approach 1:
The patent applies parameter changes by dynamically adjusting the noise level parameter in synthetic image generation. Instead of using fixed noise levels, the system varies the noise amount parameter to create a distribution of synthetic images that matches the target noise characteristics. This allows the training data to effectively represent different noise conditions without requiring precise manual control or knowledge of the actual noise amounts in the medical images.
Data Source
AI summary
A learner, a noise reduction device, and a program that can train a noise reducer even in a case where learning data having a smaller amount of noise than a normal captured image cannot be acquired or in a case where an amount of noise in learning data having the same noise as the normal captured image is unknown are provided. A learner includes a noise reducer that reduces noise of first learning data using a noise reducer that is trainable, a noise addition unit that adds predetermined noise to be added to the first learning data in which the noise is reduced, and a learning unit that trains the noise reducer to bring a distribution of the first learning data to which the noise to be added is added close to a distribution of second learning data.