A deep learning-based sparse excitation magnetoacoustic electromagnetic particle image reconstruction method

CN122289461APending Publication Date: 2026-06-26LIAONING TECHNICAL UNIVERSITY
View PDF 0 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
LIAONING TECHNICAL UNIVERSITY
Filing Date
2026-04-15
Publication Date
2026-06-26

AI Technical Summary

Technical Problem

Existing magnetoacoustic electromagnetic particle imaging technology suffers from pathological behavior during image reconstruction under sparse excitation, resulting in severe artifacts and unsatisfactory imaging effects, which limits the application of rapid imaging.

Method used

A deep learning-based sparse-excitation magneto-electromagnetic particle image reconstruction method is adopted. By constructing an improved ConvIR network and combining the system matrix under dense and sparse excitation, the network is trained using a supervised learning dataset. A multi-dimensional attention layer and feature fusion module are introduced. Pixel reconstruction loss, frequency domain constraint loss and multi-scale fusion loss are used to improve the image reconstruction quality.

Benefits of technology

While reducing the number of ultrasonic excitations, it improves image reconstruction accuracy, enhances global spatial feature extraction capabilities, suppresses artifacts, and achieves high-quality rapid imaging.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN122289461A_ABST
    Figure CN122289461A_ABST
Patent Text Reader

Abstract

This invention belongs to the field of image reconstruction technology and discloses a deep learning-based method for sparse-excitation magnetoacoustic electromagnetic particle image reconstruction. The method first applies dense and sparse excitations to the reconstruction region and acquires electrode detection signals. Dense excitation is used to obtain reference data, while sparse excitation is used to create undersampling conditions. A system matrix is ​​constructed based on the acquired signals, and a preliminary reconstructed image is obtained by solving the matrix equations. The preliminary reconstruction results are then repeatedly acquired under different magnetic particle distribution models and sparse excitation conditions. The dense excitation results are used as labels, and the sparse excitation results are used as training data to construct a supervised learning dataset. An improved ConvIR network is used for training, enabling the network to learn the image features of the sparse excitation reconstruction results and generate high-quality magnetoacoustic electromagnetic images. This method can significantly improve the image reconstruction quality of sparse-excitation magnetoacoustic electromagnetic particles and has significant research value and application prospects.
Need to check novelty before this filing date? Find Prior Art