Three Dimensional Microwave Data Inversion in Feature Space for Stroke Imaging.

Area of Science:

• Medical Imaging
• Biophysics
• Computational Electromagnetics

Background:

• Microwave imaging offers a portable, non-invasive approach for brain stroke detection.
• Conventional methods struggle with low structural detail and high computational demands.

Purpose of the Study:

• To develop an advanced microwave imaging technique for reconstructing 3D brain electrical properties.
• To enhance diagnostic accuracy and efficiency for brain stroke monitoring.

Main Methods:

• A variational autoencoder (VAE) was employed to reconstruct electrical properties in a feature space.
• The VAE decoder, incorporating prior brain knowledge, acts as a data inversion module.
• Optimization of latent codes minimized the discrepancy between measured and simulated microwave data.

Main Results:

• The proposed method achieved a 14% increase in structural similarity compared to traditional voxel-based approaches.
• Imaging speed was accelerated by over three orders of magnitude.
• The number of unknowns was reduced to only 4.8% of that used in voxel-based methods.

Conclusions:

• This feature-space reconstruction significantly enhances the resolution and efficiency of microwave brain imaging.
• The method promises more accurate stroke diagnosis and deeper insights into brain studies.