Fluoroscopic Image Noise Reduction via Curvelet Transform
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current noise reduction methods in fluoroscopic X-ray images, such as wavelet transforms, struggle to effectively reduce noise while preserving image quality and maintaining the representation of diverse orientations, leading to increased artifacts and limited radiation dose reduction.
Innovation Solution
A method utilizing discrete curvelet transforms to detect features in fluoroscopic images, match them temporally, and apply adaptive filtering to reduce noise while preserving moving objects, using gain coefficients and a cost function for feature matching and filtering.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If wavelet transform is used for noise reduction, then noise is reduced, but image contours are blurred and diverse orientations are not properly represented
Solution Approach 1:
The patent changes the mathematical transform parameters from wavelet to curvelet transform, which uses curved sinusoidal functions instead of wavelets. This parameter change enables the system to represent contours and edges more accurately while maintaining noise reduction capabilities, directly resolving the contradiction between noise reduction and contour preservation
Solution Approach 2:
The patent introduces an additional dimensional aspect by using multi-scale curvelet decomposition with multiple orientations at each scale. This dimensional expansion allows the transform to capture contours in various directions simultaneously, improving both noise reduction and contour representation compared to traditional wavelet transforms
2Object-affected harmful factors
If radiation dose is decreased to reduce patient and personnel exposure, then radiation safety is improved, but noise increases and contrast decreases
Solution Approach 1:
The patent applies temporal filtering that uses feedback from previous frames to reduce noise in current frames. By comparing and filtering signals across multiple temporal frames, the system can maintain image quality at lower radiation doses, as the feedback mechanism allows reconstruction of clear images from noisier low-dose acquisitions
Solution Approach 2:
The patent performs preliminary noise reduction through curvelet transform and temporal filtering before final image reconstruction. This preliminary processing of multiple frames allows the system to achieve acceptable image quality at lower radiation doses by pre-processing the noisy signals to extract useful information
3Measurement precision
If temporal filtering is applied to reduce noise in image sequences, then noise is reduced, but moving objects are not properly preserved
Solution Approach 1:
The patent introduces dynamic adaptation in the temporal filtering process by adjusting filter strength based on detected motion. The system dynamically modifies filtering parameters for different regions and time points, allowing strong noise reduction in static areas while preserving moving objects, thus resolving the contradiction between noise reduction and motion preservation
4Productivity
If simple temporal filtering is used to reduce computational complexity, then processing speed is improved, but artifacts remain and noise reduction is limited
Solution Approach 1:
The patent segments the image processing into distinct stages: curvelet transform decomposition, temporal filtering, and inverse transform reconstruction. This segmentation allows each stage to be optimized independently, achieving both computational efficiency and superior noise reduction performance by breaking down the complex processing into manageable segments
Data Source
AI summary
The invention relates to a method for reducing noise in fluoroscopic images by detecting traits in the curvelet domain, matching the detected traits and performing temporal filtering adapted to the type of coefficients associated with a trait. The method uses discrete curvelet transforms of the images in a sequence. The denoised coefficients are detected (20), then sent to a step of locally matching the traits (21), then the matched data undergo a temporal filtering step (22) making it possible to preserve the movement of the objects in the image and a step (23) for a 2D IDCT (Inverse Discrete Curvelet Transform) to produce the final image.