An Open-Access Experimental Dataset for Breast Microwave Imaging
Tyson Reimer, Jordan Krenkevich, Stephen Pistorius
Abstract
Microwave imaging has shown potential for breast cancer screening, but further evaluation of the clinical viability of breast microwave imaging (BMI) systems is required. Previous phantom studies have shown promise, but after decades of BMI research, simulation studies still dominate. This work addresses the challenges of small sample sizes and a lack of experimental data by providing an open-source experimental dataset, obtained using a pre-clinical BMI system. The University of Manitoba BMI Dataset (UM-BMID) contains data from 1257 phantom scans. UM-BMID is publicly available, and the community is encouraged to use it for large-scale BMI analysis. The application of logistic regression for tumor-detection on a subset of the dataset was studied to demonstrate one use of UM-BMID. The diagnostic accuracy of the classifier was (85 ± 4)%, demonstrating the promise of machine learning methods for tumor-detection in BMI.