Breast cancer is the most commonly diagnosed cancer and the leading cause of cancer death for women in Canada. Access to breast cancer screening is limited in northern and First Nation communities in Canada, as well as low- and middle-income countries (LMIC) in the broader international community. Lack of accessible screening tools contribute to inequitable care and increased mortality rates. Microwave breast detection has shown promising results as a safe, low-cost breast cancer screening method. This thesis aims to design and simulate a portable microwave device suitable for use in remote and low-income areas. The device features a cylindrical array of patch antennas and, through machine learning, will require no trained personnel to operate or obtain a preliminary diagnosis. Tiny vector network analyzers were used to measure S-parameters from 0.7 - 3 GHz. A basic radar model was improved to incorporate spherical spreading, microwave attenuation, and secondary scattering. This radar model was used to simulate time-domain sinograms of dual rod models. The rod models consisted of two point scatterers with assigned reflectivities of 100% & m% or m% & m%, where m was varied as: 10%, 30%, 50%, 70%, and 90%. The simulated sinograms were used to train a convolutional neural network to locate the rod responses. The network was able to easily detect and locate the 100% rod response, but struggled with the lower 10% and 30% reflectivities. The 100% and 90% rod models performed with an accuracy of 92%. The use of machine learning improved upon conventional reconstruction methods for breast microwave radar. Similar machine learning techniques can be performed on real scans with shell-based breast phantoms. The radar model simulations can be used to increase the data set size and improve training performance through transfer learning. Microwave sensing and machine learning methods offer promising possibilities to breast cancer screening accessibility for women in remote and low-income areas.