Deep Neural Network MRI Reconstruction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current Magnetic Resonance Imaging (MRI) technologies face challenges in reducing overall scanning time due to slow phase encoding speeds, limiting their efficiency and patient experience, especially in scenarios with limited scanning time such as abdominal scanning.
Innovation Solution
The integration of a Deep Neural Network (DNN) with accelerated imaging methods, such as parallel imaging, compressed sensing, and half-Fourier imaging, to process undersampled MRI data, allowing for the reconstruction of high-quality images from reduced k-space data, thereby shortening scanning time.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional MRI scanning with full sampling is used, then image quality is maintained, but scanning time is excessively long
Solution Approach 1:
The system performs preliminary actions by collecting calibration data during a separate calibration step, storing it in memory. This pre-collected data is then reused during actual scanning to enable accelerated imaging without sacrificing image quality, effectively preparing the system in advance to overcome the time-quality tradeoff
Solution Approach 2:
The patent introduces an intermediary computational process that uses the stored calibration data to facilitate the reconstruction of high-quality images from undersampled k-space data. This intermediary mechanism bridges the gap between reduced sampling and maintained image quality
2Productivity
If undersampling is applied to reduce scanning time, then scanning speed increases, but image quality deteriorates
Solution Approach 1:
The system incorporates feedback mechanisms where the calibration data obtained from preliminary scanning is fed into the reconstruction process. This feedback loop allows the system to compensate for the effects of undersampling, maintaining image quality while achieving faster scanning speeds
Solution Approach 2:
The patent changes key parameters by separating the calibration phase from the scanning phase, allowing the system to operate with different sampling rates at different stages. The calibration data captures system characteristics that are then used to adjust and maintain image quality during accelerated scanning
3Productivity
If calibration data is collected and stored in memory, then accelerated imaging becomes possible, but device complexity increases
Solution Approach 1:
The calibration step is performed as a preliminary action during system setup or initial operation, collecting and storing necessary calibration data in memory. This one-time preliminary action enables subsequent accelerated scanning without adding ongoing complexity to the scanning process itself
Solution Approach 2:
The system serves itself by using its own calibration data to enable accelerated imaging. The stored calibration information is reused automatically during scanning, allowing the system to improve its own performance without requiring external intervention or additional hardware complexity
Data Source
AI summary
Methods, devices, apparatus and systems for magnetic resonance imaging with deep neural networks are provided. In one aspect, a method of magnetic resonance imaging method combines a deep neural network and an accelerated imaging manner. The method includes: scanning a subject with a first undersampling factor and a first sampling trajectory to obtain first imaging information, processing the first imaging information with the deep neural network to obtain second imaging information corresponding to a second undersampling factor that is smaller than the first undersampling factor, and reconstructing a magnetic resonance image of the subject from the second imaging information.