MRI Image Normalization for CNN Noise Reduction Across Signal Levels
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing CNN-based noise reduction methods for medical images fail to effectively reduce noise when pixel values in the region of interest deviate from the pixel value range of the training images, leading to inadequate denoising.
Innovation Solution
A normalization process is applied to input images using noise and signal level information to align pixel values within the range of training images, enabling effective noise reduction with a CNN trained on normalized images.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of manufacture
If pixel values are normalized with a maximum value, then the normalization process is simple, but the pixel values in the region of interest may deviate from the training image range, causing inadequate noise reduction
Solution Approach 1:
The patent changes the normalization parameter from a fixed maximum value to a dynamic normalization factor calculated from noise information and signal level information. This allows the normalization to adapt to different noise levels and signal characteristics, ensuring pixel values fall within the training image range while maintaining reliability across various imaging conditions
Solution Approach 2:
The patent introduces feedback by using noise information and signal level information from the input image to dynamically determine the normalization factor. This feedback mechanism ensures that the normalization process adapts to the specific characteristics of each input image, improving both the applicability and effectiveness of noise reduction
2Productivity
If a fixed normalization method is used for all images, then the processing is efficient, but it cannot adapt to different noise levels and signal characteristics
Solution Approach 1:
The patent transforms the static normalization approach into a dynamic one by calculating the normalization factor based on noise information and signal level information specific to each input image. This dynamic adaptation maintains processing efficiency while significantly improving versatility across different noise levels and signal characteristics
Applied Scientific Principles
This section explains which scientific principles are used to turn an abstract innovation direction into a practical engineering solution.
Function Achieved in This Case
This approach ensures reliable noise reduction across various noise levels by normalizing input images to match the pixel value range of the training data, maintaining effective denoising performance.
Implementation Method 1
a measurement unit configured to measure nuclear magnetic resonance signals generated from a subject
Data Source
AI summary
In an image noise reduction process using a CNN, noise can be reduced effectively irrespective of signal levels and noise levels of the image. Noise information and signal level information are calculated from an image inputted in the CNN. Using the calculated information, a normalization factor suitable for the CNN is determined, and normalization of the input image is performed. The noise information is estimated from magnitude of background noise of the input image in the case where the input image is an MRI image. The signal information can be calculated, for example, as a mean value of pixel values of a subject region with respect to an image obtained by dividing the input image by the magnitude of the background noise.