Adversarial Network Defective Pixel Correction in Medical Imaging
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Defective pixels in flat panel detectors used for digital medical imaging can result in ring artifacts in reconstructed images, leading to compromised image quality, and existing methods for calibration and correction are either time-consuming or prone to misidentification and interpolation errors.
Innovation Solution
A framework utilizing adversarial networks for pixel correction, which includes a corrector and a classifier trained on datasets to identify and correct defective pixels, ensuring higher-order structure preservation and enabling online correction of defective pixels in real-time.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If traditional calibration methods are used to identify and correct defective pixels, then defective pixels can be identified and corrected, but the process is time-consuming and requires extensive onsite calibration
Solution Approach 1:
The system performs preliminary calibration offline to create a defective pixel map before actual imaging. This pre-characterization of detector defects eliminates the need for time-consuming onsite calibration, as the system can automatically apply corrections during routine operations using the pre-generated correction maps.
Solution Approach 2:
The system creates a digital copy of the defective pixel locations in the form of a defective pixel map, which is then used to correct images without requiring physical inspection or calibration during operation. This digital representation allows rapid application of corrections across multiple images.
2Reliability
If after-the-fact processing is used to correct pixels in reconstructed images, then defective pixels can be corrected, but there is a risk of misidentifying artifacts and interpolating larger areas of the image
Solution Approach 1:
The system identifies and corrects defective pixels in the raw detector data before image reconstruction occurs. By addressing defects at the source rather than in the reconstructed image, the system avoids misidentifying anatomical structures as artifacts and prevents unnecessary interpolation that would degrade image quality.
Solution Approach 2:
The system extracts and corrects defective pixel values from the raw detector signal before reconstruction. This separation of defect correction from the reconstruction process ensures that only actual defective pixels are modified, preserving the integrity of the reconstructed image and avoiding artifacts.
3Measurement precision
If extensive calibration is performed for each detector to account for failing elements, then defective pixels can be identified, but the process becomes too time-consuming for practical use
Solution Approach 1:
The system creates a digital defective pixel map that serves as a reusable correction template for multiple detectors and imaging sessions. This digital copy allows rapid deployment of correction algorithms without repeating the calibration process for each detector, significantly improving throughput while maintaining identification accuracy.
Solution Approach 2:
The calibration system is designed to generate universal correction maps that can be applied across multiple detectors and imaging conditions. This multi-functional approach allows a single calibration process to serve multiple purposes and detectors, eliminating the need for repeated calibration and improving overall productivity.
Data Source
AI summary
A framework for defective pixel correction using adversarial networks. In accordance with one aspect, the framework receives first and second training image datasets. The framework performs adversarial training of a corrector and a classifier with the first and second training image datasets respectively. The corrector may be trained to correct a first input image and the classifier may be trained to recognize whether a second input image is real or generated by the corrector. The framework applies the trained corrector to a current image to correct any defective pixels and generate a corrected image. The corrected image may then be presented.