Learnt Microwave Image Reconstruction with A Conformal Antenna Array

TL;DR

This paper introduces a deep learning approach for microwave breast imaging that uses a conformal antenna array adapting to patient-specific sizes, improving data collection and image quality.

Contribution

It integrates antenna positioning into the neural network pipeline, enabling adaptive conformal arrays for better breast image reconstruction.

Findings

The model can reconstruct breast images with good quality.

A conformal antenna array reduces signal attenuation and improves data collection.

Antenna position information enhances the neural network's focus on the region of interest.

Abstract

A deep learning model is proposed for reconstructing 2D dielectric breast images from time-domain signals. Unlike existing learning models that employ a fixed antenna array, where input data consists solely of measurements, the proposed system integrates antenna positioning into the processing pipeline. This allows for a conformal antenna array that adapts to different breast sizes for optimal data collection across various patients, which eliminates undesired signal attenuation in coupling liquid when implemented for the fixed array. By leveraging antenna positions, the breast surface can be pre-estimated, enabling the neural network to focus on image reconstruction within the region of interest. Numerical results demonstrate that the proposed model may reconstruct breast images with good quality.