Stochastic Gradient Descent and Frequency Sampling in a Radar Reconstruction Algorithm for Breast Microwave Imaging

Breast microwave imaging is an emerging diagnostic tool, but improved image reconstruction methods are needed to improve the diagnostic accuracy of the technique. This article applies stochastic gradient descent (SGD) to the optimization-based radar reconstruction (ORR) algorithm, which was previously demonstrated to improve diagnostic sensitivity and specificity compared to literature standard methods. SGD was found to reduce the computation time required to produce a reconstruction by nearly an order of magnitude and improved the reconstruction quality with respect to the loss function. This article also examines the role of frequency sampling in radar-based image reconstruction. Instead of a definite minimum frequency resolution, we propose an empirical selection that balances computational needs with data quality.