Few-view CT Reconstruction via WGAN and Stationary C-Arm
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current CT scanners face challenges with few-view CT image reconstruction due to under-sampled data, leading to streak artifacts and reduced image quality, and are often inaccessible due to their size and energy consumption.
Innovation Solution
A deep learning-based system utilizing a Wasserstein generative adversarial network (WGAN) with a generator network and discriminator network for efficient end-to-end reconstruction, learning a filtered back-projection operation in a point-wise manner to reconstruct images from few-view CT data, reducing memory burden and improving image quality.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of energy
If few-view CT reconstruction is performed with under-sampled data, then radiation exposure is reduced, but image quality deteriorates due to streak artifacts
Solution Approach 1:
The patent replaces traditional mechanical CT scanning systems with rotating gantries and multiple views with a simplified stationary C-arm geometry that acquires projections at only two angles (0 and pi). This mechanical substitution reduces radiation exposure while using deep learning algorithms to compensate for the reduced sampling, thereby maintaining image quality despite fewer views.
Solution Approach 2:
The invention changes the sampling parameters by acquiring projections at only two angles instead of multiple angles around the object. This parameter change reduces radiation exposure, and the patent compensates for the resulting image quality degradation through deep learning-based reconstruction algorithms that learn to reconstruct high-quality images from these limited angular samples.
2Manufacturing precision
If traditional CT scanners with rotating gantry are used, then image quality is maintained, but device size and energy consumption increase
Solution Approach 1:
The patent eliminates the rotating gantry mechanism entirely, replacing it with a stationary C-arm geometry that acquires projections at two fixed angles. This mechanical substitution removes the energy-consuming rotation mechanism while using computational methods (deep learning) to maintain image quality, thereby reducing energy consumption without sacrificing diagnostic capability.
Solution Approach 2:
The invention extracts and removes the rotating gantry component from the traditional CT scanner system, retaining only the essential function of acquiring projection data at two angles. This extraction eliminates the energy-consuming mechanical rotation while preserving the core imaging function through intelligent algorithms.
3Productivity
If filtered back-projection is used for reconstruction, then computational speed is improved, but image quality deteriorates with under-sampled data due to streak artifacts
Solution Approach 1:
The patent replaces the traditional filtered back-projection algorithm with a deep learning-based reconstruction algorithm. While deep learning requires more computational resources, it effectively handles under-sampled data by learning to reconstruct high-quality images from limited projections, eliminating the streak artifacts that plague filtered back-projection in few-view scenarios.
Solution Approach 2:
The invention introduces a deep learning algorithm as an intermediary between the raw projection data and the final reconstructed image. This intermediary learns to compensate for the insufficient sampling by inferring missing information, thereby bridging the gap between limited data acquisition and high-quality image reconstruction.
Data Source
AI summary
A system for few-view computed tomography (CT) image reconstruction is described. The system includes a preprocessing module, a first generator network, and a discriminator network. The preprocessing module is configured to apply a ramp filter to an input sinogram to yield a filtered sinogram. The first generator network is configured to receive the filtered sinogram, to learn a filtered back-projection operation and to provide a first reconstructed image as output. The first reconstructed image corresponds to the input sinogram. The discriminator network is configured to determine whether a received image corresponds to the first reconstructed image or a corresponding ground truth image. The generator network and the discriminator network correspond to a Wasserstein generative adversarial network (WGAN). The WGAN is optimized using an objective function based, at least in part, on a Wasserstein distance and based, at least in part, on a gradient penalty.