X-ray Photon-Counting Data Correction via Deep Learning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
X-ray photon counting detectors (PCDs) face limitations due to pulse pileup and charge splitting effects, which lead to spectral distortion and reduced counting rates, particularly in high flux environments like clinical computed tomography, affecting image quality and radiation dose management.
Innovation Solution
A deep learning-based method using a Wasserstein generative adversarial network (WGAN) framework trains artificial neural networks to correct spectral projection data for pulse pileup and charge splitting distortions, employing a pulse pileup correction ANN and a charge splitting correction ANN to generate intermediate estimates and reduce or eliminate these effects.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If x-ray photon counting detectors are used in high flux environments, then spectral dimension and material decomposition capability are improved, but pulse pileup effect causes count loss and spectral distortion
Solution Approach 1:
The patent segments the correction process into two distinct neural networks: one for pulse pileup correction and another for charge splitting correction. This segmentation allows each network to specialize in correcting specific distortion types, improving overall correction accuracy while maintaining computational efficiency.
Solution Approach 2:
The patent applies preliminary action by using deep learning models to predict and correct pulse pileup and charge splitting effects before final spectral analysis. The neural networks pre-process the distorted spectral data to restore accurate photon count information, enabling reliable material decomposition even in high flux environments.
2Measurement precision
If smaller detector pixels are used, then spatial resolution is improved, but charge splitting effect increases causing spectral distortion
Solution Approach 1:
The patent introduces deep learning models as intermediary components between the detector and the spectral analysis system. These models act as mediators that interpret the distorted signals from small-pixel detectors, separating true spectral information from charge splitting artifacts, thereby preserving spectral information while maintaining high spatial resolution.
Solution Approach 2:
The patent replaces traditional mechanical or algorithmic correction methods with data-driven deep learning models. The neural networks learn complex charge splitting patterns from training data and automatically correct spectral distortions, providing more accurate correction than conventional mechanical adjustment or simple algorithmic approaches.
3Measurement precision
If deep learning correction is applied, then spectral distortion is reduced, but computational complexity and training requirements increase
Solution Approach 1:
The patent segments the complex correction task into two specialized neural networks, making the overall system more manageable. Each network focuses on a specific distortion type, reducing the complexity of individual models while achieving comprehensive correction through their combined operation.
Data Source
AI summary
A method for x-ray photon-counting data correction. The method includes generating, by a training data generation module, training input spectral projection data based, at least in part, on a reference spectral projection data. The training input spectral projection data includes at least one of a pulse pileup distortion, a charge splitting distortion, and/or noise. The method further includes training, by a training module, a data correction artificial neural network (ANN) based, at least in part, on training data. The data correction ANN includes a pulse pileup correction ANN, and a charge splitting correction ANN. The training data includes the training input spectral projection data and the reference spectral projection data.