Measurement and Analysis of In Vivo Microwave Dielectric Properties Collected From Normal, Benign, and Malignant Rat Breast Tissues: Classification Using Supervised Machine Learning Algorithms

This work presents large-scale measurements of in vivo rat breast tissue dielectric properties (DPs) from 0.5 to 6 GHz and classifies the collected data using supervised machine learning (ML) algorithms. The main goals of this work are, first, to report in vivo animal tissue DPs for microwave medical device development and, second, to demonstrate that microwave devices can be utilized for diagnostics. To this end, we separated 18 Sprague-Dawley female rats into control and experimental groups. The experimental group was subjected to chemically induced breast cancer, and the DPs of normal tissues (NTs) and tumor tissues from the control and experimental groups were measured using the open-ended coaxial probe (OECP) technique. DPs of rat breast tissues are presented with Cole-Cole parameters. The OECP method is preferred since it can collect broadband measurements without sample preparation. Due to these advantages, the method was previously envisioned as a diagnostic tool to aid in biopsy procedures. However, high measurement error prevented the specialized device’s development and, consequently, the clinical deployment of OECP. We demonstrate that high error rates can be mitigated with the application of ML algorithms. Among seven different ML algorithms, the support vector machines (SVMs) algorithm classifies rat malignant tissues, benign tissues, and NTs with a median accuracy (ACC) of 94.4%, Matthews correlation coefficient (MCC) of 91.9%, recall of 94.4%, precision of 94.9%, and $F1$ score of 94.4%.