One-dimensional convolutional neural network (1D-CNN) image reconstruction for electrical impedance tomography

Area of Science:

• Medical Imaging
• Computational Electromagnetics
• Machine Learning

Background:

• Neural network algorithms show promise in Electrical Impedance Tomography (EIT) for improved imaging accuracy.
• Existing neural network approaches for EIT require better generalization across simulation and experimental data.
• The inverse problem in EIT image reconstruction necessitates advanced algorithmic solutions.

Purpose of the Study:

• To propose a novel one-dimensional convolutional neural network (1D-CNN) for solving the EIT inverse problem.
• To enhance the edge-preservation capabilities of reconstructed EIT images.
• To validate the anti-noise and generalization performance of the proposed 1D-CNN model.

Main Methods:

• A 1D-CNN model was developed, leveraging the characteristics of voltage data in EIT.
• Numerical simulations were used to generate abundant training samples, focusing on edge preservation.
• The network was trained and optimized using TensorFlow with an Adam optimizer.
• Reconstruction results were compared against Deep Neural Network (DNN) and 2D-CNN models.

Main Results:

• The proposed 1D-CNN demonstrated superior effectiveness and edge-preservation compared to DNN and 2D-CNN.
• The 1D-CNN exhibited enhanced anti-noise and generalization capabilities.
• Experimental validation using an EIT system confirmed the practicability of the 1D-CNN.
• The average image correlation coefficient improved by 0.0320 over DNN and 0.0616 over 2D-CNN.

Conclusions:

• The proposed 1D-CNN offers improved EIT image reconstruction, particularly for complex geometries.
• The method provides better accuracy and edge preservation than existing DNN and 2D-CNN approaches.
• The 1D-CNN shows significant potential for practical EIT applications due to its robustness and generalization.