Electrical Impedance Tomography-Based Abdominal Subcutaneous Fat Estimation Method Using Deep Learning.

Area of Science:

• Biomedical Engineering
• Medical Imaging
• Machine Learning

Background:

• Electrical Impedance Tomography (EIT) is explored for non-invasive abdominal subcutaneous fat thickness estimation.
• Traditional EIT reconstruction methods struggle with nonlinearity and ill-posedness, limiting absolute imaging accuracy.
• Accurate fat thickness measurement is crucial for metabolic and cardiovascular health assessments.

Purpose of the Study:

• To develop a deep learning-based EIT method for precise abdominal subcutaneous fat thickness estimation.
• To address the challenges of nonlinearity and ill-posedness inherent in absolute EIT imaging.
• To improve the reliability and accuracy of non-invasive body composition analysis.

Main Methods:

• A deep learning model was designed to reconstruct conductivity distributions from EIT current-to-voltage data.
• Prior anatomical information was incorporated to constrain solutions within an admissible set.
• A specialized training dataset was created to mitigate the ill-posed nature of the absolute EIT problem.
• Preprocessing involved normalizing data to reduce artifacts from electrodeposition and body geometry.

Main Results:

• The proposed deep learning method demonstrated effective estimation of abdominal subcutaneous fat thickness.
• Numerical simulations and phantom experiments validated the method's performance.
• The approach successfully addressed the nonlinearity and ill-posedness challenges in absolute EIT.
• Data normalization techniques improved the robustness of the EIT measurements.

Conclusions:

• Deep learning offers a promising solution for accurate abdominal subcutaneous fat estimation using EIT.
• The method provides a non-invasive and potentially more accessible alternative to existing techniques.
• Further research can explore clinical applications and system scalability.