Magnetic Particle Imaging Reconstruction via Time-Frequency Spectrum Denoising
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current magnetic particle imaging (MPI) technologies face challenges in efficiently removing harmonic interference and Gaussian noise simultaneously, leading to low precision in reconstructed images, with existing methods either requiring hardware upgrades or being computationally costly and non-robust.
Innovation Solution
A system and method utilizing a deep neural network (DNN) with a self-attention mechanism to perform short-time Fourier transforms, denoise time-frequency spectra, and reconstruct high-quality magnetic particle distribution models, combining global and local features to distinguish particle signals from background noise without increasing hardware complexity or time cost.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If hardware upgrades are implemented to remove background noise, then noise removal capability is improved, but system complexity and operational difficulty increase
Solution Approach 1:
The patent replaces hardware-based noise removal mechanisms with a software-based deep neural network algorithm. The DNN processes the time-frequency spectrum to remove harmonic interference and Gaussian noise, substituting complex hardware modifications with an intelligent computational approach that maintains system simplicity while improving noise removal capability.
2Reliability
If system matrix algorithms are used for signal processing, then noise removal is improved, but calculation time and computational cost increase
Solution Approach 1:
The patent substitutes traditional system matrix algorithms with a pre-trained deep neural network that operates in the time-frequency domain. The DNN processes the spectrum data through efficient forward propagation, avoiding the iterative computational processes of system matrix methods. This substitution dramatically reduces calculation time while maintaining robust noise removal performance for both harmonic interference and Gaussian noise.
3Productivity
If X-space algorithms are used for noise removal, then processing speed is improved, but robustness and effectiveness against multiple noise types decrease
Solution Approach 1:
The patent transforms the signal processing from the traditional one-dimensional time domain or frequency domain to a two-dimensional time-frequency spectrum domain. By performing short-time Fourier transform to create the spectrum, then applying the DNN in this extended dimension, the method achieves both high processing speed and robust removal of multiple noise types simultaneously. The time-frequency representation provides additional discriminatory information that enhances noise removal effectiveness while maintaining computational efficiency.
Data Source
AI summary
A method and system for reconstructing a magnetic particle distribution model based on time-frequency spectrum enhancement are provided. The method includes: scanning, by a magnetic particle imaging (MPI) device, a scan target to acquire a one-dimensional time-domain signal of the scan target; performing short-time Fourier transform to acquire a time-frequency spectrum; acquiring, by a deep neural network (DNN) fused with a self-attention mechanism, a denoised time-frequency spectrum; acquiring a high-quality magnetic particle time-domain signal; and reconstructing a magnetic particle distribution model. The method learns global and local information in the time-frequency spectrum through the DNN fused with the self-attention mechanism, thereby learning a relationship between different harmonics to distinguish between a particle signal and a noise signal. The method combines the global and local information to complete denoising of the time-frequency spectrum, thereby acquiring the high-quality magnetic particle time-domain signal.