No abstract

No abstract

A deep learning technique to enhance 3D images of the complex-valued permittivity of the breast obtained via microwave imaging is investigated. The developed technique is an extension of one created to enhance 2D images. We employ a 3D Convolutional Neural Network, based on the U-Net architecture, that takes in 3D images obtained using the Contrast-Source Inversion (CSI) method and attempts to produce the true 3D image of the permittivity. The training set consists of 3D CSI images, along with the true numerical phantom images from which the microwave scattered field utilized to create the CSI reconstructions was synthetically generated. Each numerical phantom varies with respect to the size, number, and location of tumors within the fibroglandular region. The reconstructed permittivity images produced by the proposed 3D U-Net show that the network is not only able to remove the artifacts that are typical of CSI reconstructions, but it also enhances the detectability of the tumors. We test the trained U-Net with 3D images obtained from experimentally collected microwave data as well as with images obtained synthetically. Significantly, the results illustrate that although the network was trained using only images obtained from synthetic data, it performed well with images obtained from both synthetic and experimental data. Quantitative evaluations are reported using Receiver Operating Characteristics (ROC) curves for the tumor detectability and RMS error for the enhancement of the reconstructions.

Electromagnetic inversion systems require that the experimental data be calibrated to the computational inversion model being used. In addition, accurate prior information provided to the inversion algorithm leads to higher-quality images. For some applications of inversion, such as stored grain imaging or geophysical inversion, known (calibration) targets cannot be easily introduced into the imaging region and the ability to determine prior information can be limited. In an attempt to solve the problem of calibrating data from such field-inversion systems, we introduce a work flow where: (1) a simple parametric physical model of the scattering background is obtained via a phaseless (magnitude only data) inversion algorithm that works on phase-corrupted, uncalibrated total-field measurements, and (2) we then use this simple physical model to generate calibration and prior information for subsequent full-data (magnitude and phase) inversion. Using an example of in-bin stored grain imaging, the inverted parameters are the grain angle of repose, grain height, and the average bulk permittivity of the grain. Using uncalibrated total-field data, we show that the proposed work flow obtains the overall structure of the grain in a bin despite the use of this raw data. We then show that the simple physical model can be used as both a calibration data set as well as the prior information about the grain target in a full-data (magnitude and phase) inversion. The use of this phaseless algorithm means we are able to remotely calibrate imaging systems, and obtain critical prior information about the imaging region without introducing a calibration target or physically measuring the imaging region in other ways.INDEX TERMS Inverse problems, calibration, microwave tomography, inverse imaging.

No abstract