High-frequency conductivity at Larmor-frequency in human brain using moving local window multilayer perceptron neural...

Area of Science:

• Medical Imaging
• Computational Electromagnetics
• Neuroscience

Background:

• Magnetic Resonance Electrical Properties Tomography (MREPT) visualizes internal conductivity using B1 phase data.
• Existing MREPT algorithms simplify Maxwell's equations, often introducing noise artifacts.
• Accurate conductivity mapping is crucial for understanding brain function and disease.

Purpose of the Study:

• To develop a deep learning model for direct visualization of high-frequency conductivity in the human brain.
• To overcome limitations of traditional MREPT reconstruction algorithms.
• To improve noise suppression and spatial resolution in conductivity mapping.

Main Methods:

• A novel Moving Local Window Multi-Layer Perceptron (MLW-MLP) neural network was designed.
• The MLW-MLP utilizes gradients and Laplacian of B1 phase data as input.
• Non-local mean filtering was integrated for noise reduction and resolution preservation.

Main Results:

• The deep learning model successfully visualized high-frequency conductivity distributions in the brain.
• The proposed method demonstrated effective noise suppression.
• Spatial resolution was maintained in the reconstructed conductivity images.

Conclusions:

• The developed MLW-MLP deep learning approach offers a robust method for MREPT.
• This technique provides accurate and noise-reduced conductivity mapping of the brain.
• The findings have potential implications for advanced neuroimaging and diagnostics.