Deep Convolutional Networks for Radiographic Noise and Detail Preservation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional noise suppression techniques in digital radiographic images fail to accurately characterize and correct noise behavior, particularly when exposure reduction is desirable, leading to poor image quality and potential compromise in diagnostic value.
Innovation Solution
A computer-implemented method using a machine learning network to generate a noise suppressed radiographic image by training on simulated low-exposure images, allowing user-adjustable noise reduction and additional enhancements like automatic angle and distance measurement.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Object-generated harmful factors
If conventional noise suppression techniques are used, then noise is reduced, but high-frequency image content such as edges and fine details is lost
Solution Approach 1:
The patent segments the image processing task into multiple frequency bands using wavelet decomposition. Different processing strategies are applied to different frequency components: noise suppression is applied to low-frequency components while high-frequency components containing edges and details are preserved with minimal processing. This segmentation allows simultaneous noise reduction and detail preservation that cannot be achieved with uniform processing.
Solution Approach 2:
The patent applies different processing qualities to different regions of the image based on local characteristics. Regions containing important diagnostic features (edges, anatomical structures) receive minimal processing to preserve detail, while homogeneous regions receive more aggressive noise suppression. This local adaptation ensures that noise reduction does not compromise the preservation of critical image information.
2Object-affected harmful factors
If X-ray exposure levels are reduced to follow ALARA principles, then radiation dose is decreased, but image quality deteriorates with excessive graininess and low contrast
Solution Approach 1:
The patent applies noise suppression processing as a preliminary step before diagnostic review. By pre-processing the low-dose image to reduce noise and enhance contrast, the diagnostician receives an optimized image that compensates for the reduced exposure. This preliminary action allows the use of lower radiation doses while maintaining diagnostic reliability, as the noise reduction and contrast enhancement prepare the image for optimal interpretation.
Solution Approach 2:
The patent converts the harmful effect of noise in low-dose images into a benefit by applying intelligent noise suppression algorithms. These algorithms distinguish between noise patterns and actual anatomical features, suppressing the former while preserving the latter. The noise that would normally degrade image quality at low doses is transformed into removable artifacts, allowing diagnostician to reliably interpret low-dose images as if they were higher quality.
3Object-generated harmful factors
If conventional noise suppression methods are used, then noise is reduced, but the processing is computationally intensive and time-consuming
Solution Approach 1:
The patent applies partial processing by focusing computational resources only on regions where noise suppression is most beneficial. Rather than uniformly processing the entire image with computationally intensive algorithms, the system identifies regions with high noise content and applies processing selectively. This partial action reduces computational load and processing time while maintaining effective noise suppression in critical areas.
Data Source
AI summary
A machine learning network is trained to generate a noise field image from a radiographic (x-ray) image. The training includes accessing a number of previously acquired radiographic images, duplicating the previously acquired radiographic images, and conditioning each of the duplicated images with simulated noise content to form a plurality of simulated low-exposure images. Each of the simulated low-exposure images is paired with its corresponding previously acquired image to form a learning pair. The machine learning network is trained to generate a noise field image using the learning pairs of images. A noise suppressed image of an object can be generated by applying a scaling factor to at least a portion of the corresponding noise field image and combining the scaled noise field image with a current captured image of the object.