Supervised Machine Learning for Breast Cancer Detection Using Microwave Imaging in the Frequency Domain

Due to its non-invasive and non-ionizing properties, microwave imaging has emerged as an efficient alternative to conventional screening techniques. This paper presents a Supervised Machine Learning (ML) framework for Breast Microwave classification where chosen features rely on measured scattering (S-parameters) in the frequency domain. An open source dataset from the university of Manitoba based on a preclinical Breast Microwave Imaging (BMI) system using breast phantoms (UM-BMID) is considered to illustrate that problem of detecting whether a tumor exists or not. The obtained results with maximum achieved accuracy of 98% highlight the relevance of the frequency domain features by comparing them to previously published results where features were chosen in the time domain and show the potential advantages of applying ML classification methods in that BMI system.