Deep Learning-based Classification of Breast Tumors using Raw Microwave Imaging Data

Abstract

Breast cancer is the leading type of malignant neoplasm disease among women worldwide. Breast screening makes extensive use of powerful techniques such as x-ray mammography, magnetic resonance imaging, and ultrasonography. While these technologies have numerous benefits, certain drawbacks such as the use of low-energy ionizing x-rays, a lack of specificity for malignant tissues, and cost, have motivated researchers to investigate novel imaging and detection modalities. Microwave imaging (MWI) has been extensively studied due to its low-cost structure and ability to perform measurements using non-ionizing electromagnetic waves. This study proposes a novel convolutional neural network (CNN) model for detecting and classifying tumor scatterers in MWI simulation data. To accomplish this, 10001 different numerical breast models with tumor scatterers of varying numbers and positions were developed, and the simulation results were derived using the synthetic aperture radar (SAR) technique. The presented CNN structure was trained using 8000 pieces of simulation data, and the remaining data were used for testing, achieving accuracy rates of 99.61% and 99.75%, respectively. The proposed model is compared to three state-of-the-art models on the same dataset in terms of classification performance. The results demonstrate that the proposed model effectively performs effectively well in detecting and classifying tumor scatterers.