Medical Image Denoising With Noise Correlation Maps
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image noise removal methods using convolutional neural networks (CNNs) fail to control smoothing strength, leading to noise retention in regions with high noise and signal smoothing in regions with low noise, resulting in reduced accuracy of noise reduction.
Innovation Solution
A medical image diagnostic apparatus that incorporates a processing circuitry using a learned model, which inputs a noise reduction target image and a noise correlation map to generate a denoise image by performing convolution with varying weighting coefficients based on the noise amount in different image regions, enhancing the accuracy of noise reduction.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If the same convolution process with fixed weighting factors is applied to the entire image, then the processing is simple and fast, but noise remains in high-noise regions and original signals are smoothed in low-noise regions
Solution Approach 1:
The patent applies dynamics by making the weighting factors variable rather than fixed. The convolution process adapts its smoothing strength dynamically based on local noise characteristics detected in different image regions, allowing the system to optimize noise reduction accuracy without sacrificing processing efficiency
Solution Approach 2:
The patent implements local quality by applying different weighting factors to different regions of the image based on their specific noise characteristics. High-noise regions receive stronger smoothing while low-noise regions maintain their original signals, ensuring optimal noise reduction accuracy for each local area
2Object-affected harmful factors
If stronger smoothing is applied to remove noise, then noise reduction effectiveness improves, but original image signals and textures are blurred
Solution Approach 1:
The patent applies different smoothing strengths to different regions based on their noise characteristics. High-noise regions receive stronger smoothing to effectively remove noise, while low-noise regions receive weaker or no smoothing to preserve original signals and textures, thus resolving the contradiction between noise removal effectiveness and image fidelity
Solution Approach 2:
The patent changes the smoothing parameter (weighting factor) based on local noise conditions. By adjusting the convolution weighting factors according to detected noise levels in different regions, the system achieves adaptive noise removal that preserves important image features while eliminating noise
Data Source
AI summary
According to one embodiment, a medical image diagnostic apparatus includes processing circuitry. The processing circuitry inputs a noise correlation map and a medical image or an intermediate image to a learned model that is functioned to generate a denoise image, in which noise of the medical image or noise of the intermediate image is reduced, based on the medical image generated based on data collected with respect to a subject or the intermediate image at a front stage for generating the medical image and the noise correlation map correlated with the noise included in the medical image or the intermediate image, and generates the denoise image, in which the noise of the medical image or the noise of the intermediate image is reduced.