MPI Reconstruction Using RecNet to Eliminate System Matrix
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current MPI reconstruction methods face challenges in obtaining a system matrix, leading to noisy and artifact-filled results, and those based on x-space yield poor-quality, low-resolution images, limiting the effectiveness of magnetic particle distribution imaging.
Innovation Solution
An MPI reconstruction method utilizing a RecNet model, comprising a domain conversion network and an improved UNet network, processes 1D MPI signals and velocity signals from field-free points to produce high-quality 2D images, avoiding the need for a system matrix and minimizing noise and artifacts.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If a reconstruction method based on a system matrix is used, then the reconstruction process can be performed systematically, but the system matrix is difficult to obtain and the reconstruction result contains noise and artifacts
Solution Approach 1:
The patent replaces the traditional system matrix-based mechanical reconstruction approach with a deep learning model (RecNet) that uses neural network layers to perform reconstruction. The model substitutes the complex system matrix acquisition and inversion process with learned patterns from training data, eliminating the need for explicit system matrix measurement and reducing noise and artifacts in results.
Solution Approach 2:
The patent performs preliminary training of the RecNet model using simulated data before actual reconstruction. The model is pre-trained on synthetic MPI signals and corresponding ground truth images, allowing it to learn the reconstruction mapping in advance. This preliminary action enables the model to perform high-quality reconstruction without requiring system matrix acquisition during actual operation.
2Ease of manufacture
If a reconstruction method based on x-space is used, then the reconstruction can be performed without a system matrix, but the reconstructed image has poor quality and low resolution
Solution Approach 1:
The patent transforms the 1D MPI signal into a 2D image representation through the RecNet model's architecture. The model processes the signal through multiple convolutional layers that progressively build spatial features, effectively adding dimensional information during the reconstruction process. This dimensional transformation enables high-resolution 2D image output from 1D input signals.
Solution Approach 2:
The patent changes the parameter representation from direct x-space coordinates to a learned feature space through neural network transformations. The RecNet model applies multiple convolutional layers with different kernels and activation functions to transform the input signal parameters into high-resolution image parameters, achieving both simplicity and high quality simultaneously.
3Manufacturing precision
If deep learning methods are used for MPI reconstruction, then high-quality images can be obtained without system matrix, but the model requires training data and computational resources
Solution Approach 1:
The patent creates a virtual copy of the MPI imaging process through simulated training data. Instead of requiring physical system matrix measurements, the model is trained on synthetically generated MPI signals and corresponding ground truth images. This copying approach allows the model to learn from idealized data without the complexity of actual system matrix acquisition and calibration.
Data Source
AI summary
An MPI reconstruction method, device, and system based on a RecNet model include obtaining a one-dimensional (1D) MPI signal on which imaging reconstruction is to be performed, taking the 1D MPI signal as an input signal, and inputting the input signal and a velocity signal of an FFP corresponding to the input signal into a trained magnetic particle reconstruction model RecNet for image reconstruction to obtain a two-dimensional (2D) MPI image, where the magnetic particle reconstruction model RecNet is constructed based on a domain conversion network and an improved UNet network. The MPI reconstruction method, device, and system obtain a high-quality and clear magnetic particle distribution image without obtaining the system matrix.