Ultrasound Image Classification of Breast Cancer Using MobileNet

Abstract

Breast cancer is one of the most prevalent diseases affecting women and has a high mortality rate if not detected at an early stage. Therefore, the development of an automated and accurate system for breast cancer diagnosis is of critical importance. One of the most commonly used methods for early breast cancer detection is medical ultrasonography (US) imaging, as it is safe and easily accessible. However, ultrasound images suffer from several limitations, including low image quality, high noise levels, and heterogeneous characteristics, which make the classification of cancer types challenging. In this study, a transfer learning approach is employed for breast ultrasound image classification by utilizing the MobileNet architecture, which is lightweight and computationally efficient, to enhance model performance. The classification task is performed on three classes: benign tumors, malignant tumors, and normal tissue. The dataset used is the BUSI (Breast Ultrasound Images) dataset obtained from Baheya Hospital, Cairo, Egypt, consisting of 780 breast ultrasound images. Experiments are conducted using several pre-trained architectures, including MobileNet, MobileNetV2, Xception, and InceptionV3. The evaluation results demonstrate that the MobileNet architecture achieves the best performance with an F1-score of 89%. These results indicate that the proposed approach is effective for classifying ultrasound images, as features are automatically and globally learned by the neural network without requiring manual geometric feature analysis.