CT Imaging Method for Truncated Data Prediction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Computed tomography (CT) images suffer from data truncation issues when objects exceed the scanning field, leading to distorted and inaccurate images due to incomplete projection data, which conventional methods fail to address effectively, resulting in artifacts and contamination of data error channels.
Innovation Solution
An imaging method that preprocesses projection data using a trained learning network to predict the truncated portion, performs forward projection on the predicted image, and reconstructs the image based on both predicted and untruncated projection data, employing deep learning techniques to improve image quality and reduce artifacts.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If conventional mathematical models (water model) are used to predict truncated portion, then the reconstruction process can be completed, but image quality varies and performance is not ideal
Solution Approach 1:
The patent changes the parameter of prediction method from conventional mathematical models (water model) to deep learning-based prediction. The deep learning model learns complex relationships between untruncated and truncated portions from training data, enabling accurate prediction of truncated region while maintaining reconstruction completeness. This parameter change resolves the contradiction by providing both reliable completion and high image quality.
2Reliability
If image reconstruction is performed using size smaller than or equal to scanning field, then reconstruction can be completed, but CT value is incorrect
Solution Approach 1:
The patent applies preliminary action by predicting the truncated portion before performing image reconstruction. The deep learning model predicts the truncated region based on untruncated data, and this predicted information is then incorporated into the reconstruction process. This preliminary prediction ensures that both truncated and untruncated regions are properly represented, achieving both reliable completion and accurate CT values.
3Area of stationary object
If data truncation occurs, then scanning field is limited, but artifacts and contamination appear in reconstructed image
Solution Approach 1:
The patent introduces an intermediary - the deep learning prediction model - that mediates between the limited scanning field data and the need for complete image reconstruction. The model processes untruncated portion data and generates predictions for truncated portions, acting as a bridge that fills in missing information without introducing artifacts. This intermediary approach eliminates contamination while working within the limited scanning field constraints.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
The present application provides an imaging method and system, and a non-transitory computer-readable storage medium. The imaging method comprises preprocessing projection data to obtain a predicted image of a truncated portion; performing forward projection on the predicted image to obtain predicted projection data of the truncated portion; and performing image reconstruction using the projection data obtained by forward projection and projection data of an untruncated portion.