Spatially-Variant PSF Neural Network for CT Resolution
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional super-resolution methods for computed tomography (CT) images fail to effectively address spatial variations in image resolution, leading to suboptimal image quality due to the neglect of the system's point spread function, which results in inefficient neural network performance and high computational demands.
Innovation Solution
A neural network framework is introduced that incorporates a spatially-variant point spread function model to enhance CT image resolution, combining super-resolution techniques with intrinsic physics-based information, reducing feature dependency on low-resolution images and enabling faster processing by integrating the PSF within the network.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Manufacturing precision
If conventional super-resolution methods are applied to CT images, then image resolution is improved, but spatial variations in resolution are not addressed leading to suboptimal image quality
Solution Approach 1:
The patent applies local quality by incorporating a spatially-variant point spread function model that adapts to different regions of the CT image. Instead of using a uniform super-resolution approach, the system models the specific blurring characteristics of each spatial location, allowing the neural network to process each region according to its local resolution properties, thereby addressing spatial variations and improving overall image quality
Solution Approach 2:
The patent applies preliminary action by integrating the point spread function model into the neural network architecture before the actual super-resolution processing occurs. The PSF characteristics are pre-computed and embedded in the network, enabling the system to account for spatial variations from the outset rather than attempting to correct them after uniform super-resolution is applied
2Manufacturing precision
If conventional super-resolution methods are used, then image resolution is enhanced, but computational demands increase
Solution Approach 1:
The patent reduces computational demands by pre-computing and integrating the point spread function model into the neural network architecture. The PSF characteristics are calculated in advance and embedded as fixed components, eliminating the need for real-time PSF computation during super-resolution processing, thereby reducing overall computational burden while maintaining accuracy
Solution Approach 2:
The patent merges the point spread function model with the super-resolution neural network into a unified architecture. By combining these two previously separate processing stages into a single integrated system, the patent eliminates redundant computations and enables more efficient processing that can be executed on GPUs
3Reliability
If spatially-variant PSF modeling is incorporated, then image quality improves, but neural network complexity increases
Solution Approach 1:
The patent reduces neural network complexity by pre-computing the spatially-variant PSF model and embedding it as fixed components within the network architecture. This preliminary preparation transforms a potentially complex adaptive model into a structured framework with predetermined parameters, making the network more manageable and efficient while retaining the ability to account for spatial variations
4Manufacturing precision
If multi-stage neural network processing is applied, then image resolution is progressively enhanced, but processing time increases
Solution Approach 1:
The patent merges multiple processing stages into a single unified neural network execution. Instead of sequentially applying separate super-resolution steps, the integrated architecture processes the image through all resolution enhancement stages in one pass, utilizing the pre-integrated PSF model to guide the entire process simultaneously, thereby reducing processing time while maintaining progressive resolution enhancement
Data Source
AI summary
The present disclosure relates to a spatially-variant model of a point spread function and its role in enhancing medical image resolution. For instance, a method of the present disclosure comprises receiving a first medical image having a first resolution, applying a neural network to the first medical image, the neural network including a first subset of layers and, subsequently, a second subset of layers, the first subset of layers of the neural network generating, from the first medical image, a second medical image having a second resolution and the second subset of layers of the neural network generating, from the second medical image, a third medical image having a third resolution, and outputting the third medical image, wherein the first resolution is lower than the second resolution and the second resolution is lower than the third resolution.