Breast cancer can be considered one of the significant causes of death, especially among women worldwide. Therefore, it is essential to detect and diagnose breast cancer as early as possible to reduce the adverse effects on patients and protect women’s health. This study proposes a model applying transfer learning and fine-tuning to classify and detect benign, malignant breast cancer, and normal breast. We train the proposed model with transfer learning from the pre-trained MobileNet model to identify breast cancers and optimize the prediction results. The dataset contains 780 ultrasound images categorized into three classes which are benign breast cancer (437 images), malignant breast cancer (210 images), and normal breast (133 images). The experimental results show that applying the transfer learning and fine-tuning technique from the MobileNet model achieves promising results, with the accuracy and F1-score being 0.9651–0.9648, 0.9412–0.9417, and 0.9060–0.9085, respectively, with three scenarios.