Automatic Breast Cancer Exploration Using Pre-Trained Deep Convolutional Neural Networks (CNNs)
Marwa Naas, Hiba Mzoughi, Ines Njeh, Mohamed Ben Slima · 2023
Breast cancer, one of the most common types of cancer among women, has seen an increase in incidence and mortality rates over the years. Its development typically takes several months or even years, underscoring the importance of early diagnosis for effective treatment and improved outcomes. In this paper, we use the concept of transfer learning, we present a novel Deep Learning (DL) framework for breast cancer classification using ultrasound (US) images. The proposed framework investigates several pre-trained Convolutional Neural Networks (CNNs) on the large-scale ImagesNet dataset, including Residual Networks101, AlexNet, Inception-ResNet-v2, and Inception_ v3. Various metrics, such as accuracy, f1_measure, Kappa, and Roc_area, have been used for performance evaluation of the studied transfer-learning approaches. Simulation results, using the publicly available BUSI dataset, reveal that ResNet 101 outperforms the other tested DL networks, achieving an overall accuracy of 0.948 over the testing dataset. Even with a small dataset, the proposed method delivers competitive results.