CT Projection Frequency Splitting for Detail Visibility
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In computed tomography (CT) scans, higher frequency details such as small blood vessels or lesions become less visible due to mixed and combined frequency components in projection data, and noise is not uniformly distributed across different organs, making them harder to detect.
Innovation Solution
A medical imaging system that decomposes projection data into frequency components covering predetermined frequency bands and applies a trained neural network model to improve image quality, using lower and higher quality training data pairs to generate processed frequency components, which are then composed to produce higher quality images.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If frequency components are mixed and combined together in projection data, then the data processing is simplified, but higher frequency details become less visible and harder to detect
Solution Approach 1:
The projection data is divided into multiple frequency components, each covering a specific frequency band. This segmentation allows the system to process and enhance different frequency ranges separately, making higher frequency details (such as small blood vessels or lesions) more visible while maintaining manageable processing complexity through structured frequency band allocation.
2Loss of information
If noise is present in projection data, then the data contains real signal information, but noise distribution varies across different organs making uniform processing difficult
Solution Approach 1:
The system applies different processing strategies to different frequency bands and regions. By segmenting the data into frequency components and applying organ-specific processing parameters, the system can adapt to the varying noise characteristics of different organs while preserving signal information. This allows tailored processing for each organ region rather than uniform processing.
3Measurement precision
If a trained neural network model is applied to process frequency components, then image quality is improved, but the processing time and computational resources increase
Solution Approach 1:
The neural network model is pre-trained using lower quality projection data as input and higher quality projection data as target output. This preliminary training allows the model to learn the transformation patterns beforehand, so that during actual processing, the model can quickly apply these learned patterns to enhance image quality without requiring extensive real-time computation for each new input.
Data Source
AI summary
Data acquired from a scan of an object can be decomposed into frequency components. The frequency components can be input into a trained model to obtain processed frequency components. These processed frequency components can be composed and used to generate a final image. The trained model can be trained, independently or dependently, using frequency components covering the same frequencies as the to-be-processed frequency components. In addition, organ specific processing can be enabled by training the trained model using image and/or projection datasets of the specific organ.