Deep Learning Reconstruction for Head-and-Neck Vessel Wall MRI
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current magnetic resonance imaging technologies face challenges in achieving high spatial resolution and short scanning times for head-and-neck integrated blood vessel wall imaging, making it difficult to efficiently recognize plaques.
Innovation Solution
A method and apparatus that transform undersampled magnetic resonance K-space data into high-resolution images using deep learning reconstruction and recognition models, specifically employing dense connection networks and residual blocks for enhanced feature propagation and plaque detection.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional magnetic resonance imaging methods are used to achieve high spatial resolution, then imaging quality is improved, but scanning time increases significantly
Solution Approach 1:
The patent applies partial sampling in the K-space domain, acquiring only a subset of the full K-space data (e.g., central region or selected lines) rather than complete sampling. This partial action reduces scanning time while deep learning reconstruction algorithms compensate for the missing data to achieve high-resolution images, effectively resolving the contradiction between scanning time and spatial resolution
Solution Approach 2:
The patent introduces deep learning reconstruction algorithms as an intermediary between the undersampled K-space data and the final high-resolution image. This intermediary component processes the incomplete data, inferring missing information through learned patterns from training data, thereby enabling high spatial resolution without requiring complete data acquisition that would extend scanning time
2Productivity
If scanning time is reduced through undersampling, then productivity is improved, but image quality and recognition accuracy deteriorate
Solution Approach 1:
The patent uses deep learning models trained on fully-sampled high-quality images to create a mapping that allows reconstruction of high-resolution images from undersampled data. The model learns to copy or replicate the quality characteristics of fully-sampled images by inferring missing information from the available undersampled data, thus maintaining image quality while improving imaging speed
Solution Approach 2:
The patent changes the sampling parameters in the K-space domain, using variable density sampling or selective sampling strategies that prioritize acquisition of critical frequency components. This parameter change allows faster imaging while the deep learning reconstruction compensates for the altered sampling pattern, maintaining image quality despite reduced sampling
3Measurement precision
If deep learning reconstruction models are used to improve image quality from undersampled data, then measurement precision is improved, but device complexity increases
Solution Approach 1:
The patent replaces traditional iterative mathematical reconstruction methods with deep learning-based reconstruction. This substitution uses trained neural networks that perform reconstruction in a single forward pass rather than through multiple iterative computations, reducing computational complexity while improving reconstruction accuracy from undersampled 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 approach enables efficient transformation of undersampled images into high-resolution images, allowing for accurate plaque recognition and improving the speed and accuracy of plaque detection in magnetic resonance imaging.
Implementation Method 1
transforming the magnetic resonance undersampled K-space data to the image domain through inverse Fourier transform to obtain the preprocessed image
Data Source
AI summary
A method for magnetic resonance imaging and plaque recognition includes: obtaining magnetic resonance undersampled K-space data; transforming the magnetic resonance undersampled K-space data to an image domain through inverse Fourier transform to obtain a preprocessed image; reconstructing the preprocessed image through a pre-established deep learning reconstruction model to obtain a high-resolution imaging image of a blood vessel wall; and recognizing plaques in the high-resolution imaging image of the blood vessel wall through a pre-established deep learning plaque recognition model. A neural network corresponding to the pre-established deep learning reconstruction model is a dense connection network. The magnetic resonance undersampled K-space data is head-and-neck combined magnetic resonance undersampled K-space data of the blood vessel wall.