GSR: A Gaussian Splatting-Based Reconstruction Framework for EIT.

Area of Science:

• Biomedical Imaging
• Computer Vision
• Inverse Problems

Background:

• Electrical Impedance Tomography (EIT) reconstructs internal conductivity distributions.
• Traditional EIT reconstruction faces challenges with accuracy and efficiency.
• 2D Gaussian Splatting (GS) excels in high-resolution image representation and rendering.

Purpose of the Study:

• To introduce and evaluate 2D Gaussian Splatting for EIT reconstruction.
• To develop a novel GS-based framework for modeling conductivity distributions.
• To improve the accuracy and efficiency of EIT image reconstruction.

Main Methods:

• Modeling conductivity distributions using Gaussian kernels with adjustable parameters (position, covariance, amplitude).
• Integrating threshold-adjusted ReLU for component filtering and Sigmoid function for conductivity constraints.
• Applying the framework to simulated and real EIT datasets.

Main Results:

• The GS-based EIT reconstruction framework demonstrates superior performance compared to traditional model-driven methods.
• Reconstruction quality is competitive with existing neural network-based approaches.
• Ablation studies validate the effectiveness of the framework's core components.

Conclusions:

• 2D Gaussian Splatting offers a promising new approach for EIT reconstruction.
• The proposed framework enhances representation accuracy and adheres to physical constraints.
• This integration bridges computer vision rendering techniques with biomedical imaging challenges.