Deep Learning applied to 3D reconstruction of Inverse Problem in Electrical Impedance Tomography

• Grant number: 25/04329-4
• Support Opportunities: Scholarships in Brazil - Doctorate (Direct)
• Start date: August 01, 2025
• End date: July 31, 2030
• Field of knowledge: Engineering - Biomedical Engineering - Bioengineering
• Principal Investigator: Marcos de Sales Guerra Tsuzuki
• Grantee: Jungeui Choi
• Host Institution: Escola Politécnica (EP). Universidade de São Paulo (USP). São Paulo , SP, Brazil

Abstract

The project proposes the use of deep neural networks for the three-dimensional reconstruction of images in Electrical Impedance Tomography (EIT), aiming to enhance the quality and efficiency of traditional methods. The approach includes automatic segmentation of medical images, generation of conductivity meshes, and the development of advanced generative models. To achieve this, various deep learning architectures will be explored, including Implicit Neural Representations (INRs), Diffusion Models (DMs), Normalizing Flow (NF), Kolmogorov-Arnold Networks (KANs), and Transformers. The networks will be trained using segmented computed tomography images converted into finite element meshes representing thoracic conductivity. Model validation will be conducted through quantitative and qualitative analyses, comparing their performance with traditional approaches. The ultimate goal is to develop a robust and generalizable solution capable of handling anatomical variations and data noise, contributing to the advancement of EIT as a medical imaging technique.