Deep Unfolding Network for Nonlinear Multi-Frequency Electrical Impedance Tomography

2025-07-22

AI Summary

To address the low reconstruction accuracy of frequency-dependent tissue conductivity and poor separability of overlapping components in nonlinear multi-frequency electrical impedance tomography (mfEIT), this paper proposes a physics-informed deep unrolling network. The method unrolls a Gauss–Newton iterative scheme—augmented with proximal regularization—into a learnable layered architecture, while incorporating a graph neural network (GNN) to explicitly model inter-frequency conductivity correlations on triangular meshes, thereby preserving geometric and physical consistency. By jointly embedding nonlinear forward modeling, variational priors, and data-driven representations, the framework ensures interpretability without compromising generalizability. Experimental results demonstrate that the proposed approach significantly improves conductivity reconstruction accuracy in complex, overlapping regions, achieving superior spatial resolution and frequency-response modeling performance compared to state-of-the-art methods.

Technology Category

• Machine Learning: Graph-based Machine Learning
• Constraint Satisfaction and Optimization: Other Foundations of Constraint Satisfaction
• Search and Optimization: Non-convex Optimization

Application Category

• Graph Algorithms and Modeling for the Web: Graph neural networks and deep learning approaches for Web-related graphs
• Systems and Infrastructure for Web, Mobile and WoT: Applied ML and AI for Web-based mobile applications
• Economics, Online Markets and Human Computation: Incentives in network design for Web infrastructures and ecosystems

Abstract

Multi-frequency Electrical Impedance Tomography (mfEIT) represents a promising biomedical imaging modality that enables the estimation of tissue conductivities across a range of frequencies. Addressing this challenge, we present a novel variational network, a model-based learning paradigm that strategically merges the advantages and interpretability of classical iterative reconstruction with the power of deep learning. This approach integrates graph neural networks (GNNs) within the iterative Proximal Regularized Gauss Newton (PRGN) framework. By unrolling the PRGN algorithm, where each iteration corresponds to a network layer, we leverage the physical insights of nonlinear model fitting alongside the GNN's capacity to capture inter-frequency correlations. Notably, the GNN architecture preserves the irregular triangular mesh structure used in the solution of the nonlinear forward model, enabling accurate reconstruction of overlapping tissue fraction concentrations.

Problem

• Estimating tissue conductivities across multiple frequencies in mfEIT
• Combining iterative reconstruction with deep learning for interpretability
• Reconstructing overlapping tissue concentrations using GNNs and PRGN framework

Innovation

• Variational network combining iterative reconstruction and deep learning
• Graph neural networks integrated within PRGN framework
• Unrolled PRGN algorithm with GNN for frequency correlations