A Machine Learning Workflow for Tumour Detection in Breasts Using 3D Microwave Imaging

Abstract

A two-stage workflow for detecting and monitoring tumors in the human breast with an inverse scattering-based technique is presented. Stage 1 involves a phaseless bulk-parameter inference neural network that recovers the geometry and permittivity of the breast fibroglandular region. The bulk parameters are used for calibration and as prior information for Stage 2, a full phase contrast source inversion of the measurement data, to detect regions of high relative complex-valued permittivity in the breast based on an assumed known overall tissue geometry. We demonstrate the ability of the workflow to recover the geometry and bulk permittivity of the different sized fibroglandular regions, and to detect and localize tumors of various sizes and locations within the breast model. Preliminary results show promise for a synthetically trained Stage 1 network to be applied to experimental data and provide quality prior information in practical imaging situations.

Machine-Learning-Enabled Recovery of Prior Information From Experimental Breast Microwave Imaging Data

“…In [33] we showed the ability of this network architecture to accurately predict p 4 (permittivity and geometry) of symmetric fibroglandular regions (Figure 3) from noisy synthetic data, and showed that the combination of adipose and fibroglandular tissues can be used as an appropriate background for tumor detection. In our previous attempts to recover the same parameters from calibrated experimental data for the same symmetric fibroglandular phantom used in this work we were able to accurately recover geometry information, but struggled to recover reasonable estimates for the permittivity of the fibroglandular tissue.…”

“…Full details on the neural network and the overall proposed workflow are presented in our previous work [33], but are summarized here for convenience.…”

“…Our recent efforts have been focused on exploring the various potential applications of machine learning for 3D image reconstruction of data from electromagnetic imaging systems. We have applied these techniques to both stored grain monitoring [15,19], and microwave breast imaging [32,33] and have demonstrated that machine-learning-assisted parametric inversion can be an efficient way to obtain prior information about the region of interest. This approach has been successfully used to calibrate experimental data to a computational model for full phase inversion.…”

“…In this paper we focus on extracting parametric prior information from relatively simple 3D experimental breast phantoms using a fully connected neural network, but consider the specific case where the physical tissues do not match our simple model well. This work is an extension of [33] which focused primarily on synthetic tests. Here we expand upon previous synthetic tests by performing parametric inversion on synthetic data where the geometry of the phantom is not modeled by the synthetic training data.…”

Microwave Imaging for Early Breast Cancer Detection: Current State, Challenges, and Future Directions

“…The advantages of CNN-based learning include improved computational efficiency, especially when solving inverse scattering problems. Examples of studies that used machine learning and deep learning neural networks to simplify the inverse problem are given in [ 102 , 118 , 119 , 120 , 121 ]. In addition, a 2D-based imaging algorithm may not work for a 3D biological object and might lead to inaccurate reconstruction results.…”