Dual-Domain Neural Network for Limited-Angle 3D Imaging Reconstruction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current C-arm x-ray systems face challenges in delivering high-quality 2D projections for 3D reconstruction due to limited rotation angles, making it difficult to perform accurate 3D visualization, especially in community hospitals, and traditional geometric calibration methods are cumbersome for real-time image reconstruction.
Innovation Solution
A dual-domain neural network is used to generate high-quality 3D volumetric images from limited angle 2D projections by integrating imaging parameters and simulating x-ray interactions with tissue, enabling real-time self-calibration and improving image quality.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of manufacture
If traditional C-arm x-ray systems are used with limited rotation angles, then the equipment is more affordable and accessible, but the image quality and 3D reconstruction accuracy deteriorate
Solution Approach 1:
The patent uses neural networks to learn and replicate the characteristics of high-quality 3D CT images from limited-angle 2D X-ray projections. The neural network models (first neural network for sinogram optimization, second neural network for 3D reconstruction) copy and generate virtual complete-angle projection data and high-quality 3D volumetric images that mimic the appearance and detail of traditional CT scans, enabling accurate 3D visualization without requiring expensive full-angle rotation equipment
Solution Approach 2:
The patent transforms the imaging approach by changing from physical parameter changes (rotation angle) to data parameter changes. Instead of physically rotating the C-arm to acquire data from all angles, the system uses neural networks to synthesize and optimize sinogram parameters and 3D reconstruction parameters, generating high-quality images from limited angular data through computational methods
2Manufacturing precision
If geometric calibration is performed for each image to ensure accurate 3D reconstruction, then the reconstruction accuracy improves, but the process becomes cumbersome and time-consuming
Solution Approach 1:
The patent implements self-calibration functionality where the system automatically determines and corrects geometric calibration parameters without manual intervention. The neural networks are trained to automatically compensate for variations in C-arm position, rotation angles, and imaging geometry, performing calibration operations autonomously during the image acquisition and reconstruction process, thereby eliminating time-consuming manual calibration procedures
Solution Approach 2:
The neural networks perform preliminary learning and optimization during the training phase, where they are pre-trained on large datasets of calibrated images and sinograms. This preliminary action enables the networks to automatically handle calibration corrections during actual surgical procedures without requiring real-time manual calibration, significantly reducing operational time
3Manufacturing precision
If 180 degrees plus fan angle rotation is achieved for complete 3D reconstruction, then the image quality improves, but the system cost and complexity increase
Solution Approach 1:
The patent uses neural networks to copy and generate virtual complete-angle projection data from limited-angle acquisitions. The first neural network optimizes sinograms by learning patterns from complete-angle CT data, and the second neural network reconstructs 3D volumetric images, effectively copying the quality and completeness of full-angle scans without requiring the physical C-arm system to rotate through the full 180-plus degree range
Solution Approach 2:
The patent replaces the mechanical rotation system with a computational reconstruction system. Instead of mechanically rotating the C-arm to acquire data from all angles, the system uses neural networks to computationally generate the necessary projection data and 3D reconstructions from limited angular views, substituting mechanical complexity with algorithmic processing
4Ease of manufacture
If traditional filtered back-projection is used for 3D reconstruction, then the reconstruction process is straightforward, but the radiation dose and data requirements increase
Solution Approach 1:
The neural networks learn to copy and reconstruct high-quality 3D images from a fraction of the data required by traditional methods. The networks are trained on complete-angle CT datasets and learn to synthesize high-quality 3D volumetric images from only 30-50% of the projection data that traditional filtered back-projection would require, thereby reducing radiation dose while maintaining image quality
Solution Approach 2:
The patent changes the reconstruction approach from data-driven (requiring large amounts of projection data) to model-driven (using learned neural network models). The neural networks incorporate prior knowledge about tissue characteristics and imaging physics, enabling high-quality reconstruction from reduced data quantities by leveraging learned patterns rather than relying solely on raw projection data
Applied Scientific Principles
This section explains which scientific principles are used to turn an abstract innovation direction into a practical engineering solution.
Function Achieved in This Case
This solution allows for accurate 3D visualization in real-time, reducing the need for expensive equipment and improving clinical efficiency, while requiring only 30-50% of the data needed for traditional 3D reconstruction, thus enhancing image quality and reducing radiation dose.
Implementation Method 1
inputting the updated sinogram into a dual-domain neural network, wherein the dual-domain neural network includes: a first neural network trained to optimize a sinogram constructed based on limited angle fluoroscopic images, and a second neural network trained to optimize a 3D reconstruction that is generated based on the sinogram optimized by the first neural network
Data Source
AI summary
This application describes a method and system for medical 3D imaging reconstruction based on limited-angle 2D image acquisition, which only requires 30˜50% of the data required by traditional scanner-based 3D reconstruction. This approach significantly reduces the radiation dose requirement and improves clinical efficiency. An example method includes: capturing a plurality of 2-dimensional (2D) projections of a target that cover a limited angle of the target; calibrating the plurality of 2D projections of the target to obtain a plurality of calibrated 2D projections; generating a sinogram based on the plurality of calibrated 2D projections; inputting the sinogram into a dual-domain neural network to obtain a 3D volumetric image of the target that is geometrically corrected and calibrated.