Neural Network Tomographic Data Estimation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Traditional interpolation and extrapolation techniques are inadequate for addressing missing or incomplete data in tomographic reconstruction, leading to image artifacts, and are often slow to compute.
Innovation Solution
The use of deep learning techniques, specifically trained neural networks, to estimate and correct missing or corrupted data in the tomographic reconstruction process, by processing scan data and generating estimated datasets to produce corrected scan data for improved image reconstruction.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If traditional interpolation or extrapolation techniques are used to estimate missing data, then the reconstruction process can be performed, but image artifacts occur and computation is slow
Solution Approach 1:
The patent transforms the missing data estimation problem from traditional mathematical interpolation/extrapolation to a deep learning parameter estimation problem. Neural networks are trained to learn the mapping between available data and missing data, changing the fundamental approach from deterministic mathematical operations to data-driven parameter learning, which simultaneously improves image quality and enables faster computation through optimized network inference
Solution Approach 2:
The patent replaces traditional mechanical/mathematical interpolation and extrapolation algorithms with a neural network-based system. Instead of using conventional mathematical operations to estimate missing data, the system uses trained neural networks that have learned patterns from complete datasets, substituting the mechanical computation process with an intelligent system that achieves both higher accuracy and faster processing
2Measurement precision
If traditional interpolation or extrapolation techniques are used to estimate missing data, then the reconstruction process can be performed, but image artifacts occur
Solution Approach 1:
The patent changes the estimation approach from traditional mathematical methods to deep learning-based parameter estimation. Neural networks are trained to predict missing data parameters with high accuracy by learning from complete datasets, fundamentally improving measurement precision and eliminating the artifacts that plague traditional interpolation and extrapolation methods
Solution Approach 2:
The patent substitutes traditional mathematical interpolation and extrapolation mechanisms with neural network-based estimation. This replacement eliminates the inherent limitations of conventional methods that produce artifacts, as the neural network learns complex patterns and relationships that traditional mathematics cannot capture, thereby improving data estimation accuracy and eliminating harmful artifacts
Data Source
AI summary
A method relates to the use of deep learning techniques, which may be implemented using trained neural networks (50), to estimate various types of missing projection or other unreconstructed data. Similarly, the method may also be employed to replace or correct corrupted or erroneous projection data as opposed to estimating missing projection data.