A Comparative Evaluation of Microimpedance Tomography Reconstruction Algorithms for in Vitro Imaging

Area of Science:

• Biomedical Engineering
• Machine Learning
• Image Reconstruction

Background:

• Electrical Impedance Tomography (EIT) is a valuable imaging technique.
• Traditional EIT reconstruction methods can be limited in accuracy and resolution.
• Developing advanced reconstruction algorithms is crucial for EIT applications.

Purpose of the Study:

• To develop a novel miniature glass EIT system.
• To train and validate an open-source machine learning model for EIT image reconstruction.
• To benchmark the AI model against established reconstruction methods.

Main Methods:

• Development of a novel miniature glass EIT system.
• Training and validation of a 1-dimensional Convolutional Neural Network (1D-CNN) for image reconstruction.
• Benchmarking against Gauss-Newton and Total Variation methods using synthetic and experimental data.

Main Results:

• The 1D-CNN model achieved up to a 5-fold reduction in mean square error compared to traditional methods on synthetic data.
• Experimental validation demonstrated superior EIT reconstruction capabilities with an average positional accuracy of 147 μm and dimensional resolution of 70 μm.
• The system successfully monitored zebrafish development, showing compatibility with biological imaging.

Conclusions:

• The developed miniature EIT system and 1D-CNN model offer significant advantages in image reconstruction accuracy and resolution.
• This AI-driven approach shows promise for various EIT applications, including in vitro biological monitoring.
• The open-source nature of the model facilitates further research and development in EIT imaging.