An Effective Framework for Deep-Learning-Enhanced Quantitative Microwave Imaging and Its Potential for Medical...

Area of Science:

• Medical Imaging
• Electromagnetics
• Artificial Intelligence

Background:

• Microwave imaging offers a promising alternative to conventional medical diagnostics.
• The inverse scattering problem in microwave imaging is challenging due to non-linearity and ill-posedness.

Purpose of the Study:

• To present an innovative, automated, and reliable solution for the inverse scattering problem in microwave imaging.
• To enhance the accuracy and applicability of microwave imaging for biomedical diagnostics, including brain stroke detection.

Main Methods:

• A two-step framework combining orthogonality sampling and deep learning (U-Net).
• Orthogonality sampling generates initial target shape and contrast estimates from scattered field data.
• A U-Net neural network performs image segmentation to refine target shape and retrieve precise contrast values.

Main Results:

• The proposed method successfully reconstructed target shapes and contrast values.
• Validation with synthetic and experimental (Fresnel database) data demonstrated reliable performance.
• A numerical example illustrated potential for microwave brain stroke diagnosis.

Conclusions:

• The combined orthogonality sampling and U-Net approach provides an effective solution to the inverse scattering problem.
• This technique shows significant potential for advancing biomedical microwave imaging applications.
• The automated nature and validated accuracy pave the way for wider adoption in medical diagnostics.