Breast cancer is a commonly occuring and serious health condition that affects cells in the breast. Early detection of breast cancer increases the survival rates and can be achieved through regular screenings using image modalities such as ultrasound and mammography. Automated solutions for detection of cancerous tissue in the breast helps radiologists make accurate and timely decisions. In addition, highlighting the most important regions in the scanned breast images that influence decision-making assist clinicians when examining the images. This research proposed a model based on Vision Transformer to classify ultrasound breast images into benign and malignant. The model was trained using the Ultrasound Breast Images for Breast Cancer dataset ( https://www.kaggle.com/datasets/vuppalaadithyasairam/ultrasound-breast-images-for-breast-cancer ). The model also integrates Grad-CAM for making visual interpretations for helping clinicians to take decisions. The proposed model obtains a 99% accuracy, 100% precision, 98% recall, and 98.99% F1-score. With the integration of Grad-CAM, the model highlights the significant regions that highly influence the decision making, providing radiologists with interpretable visual patches to support their clinical assessments. Additionally, we employed EfficientNetB0 and MobileNetV2 for evaluating the performance of our proposed architecture, and results show that our model outperforms these pre-trained models.