Ultrasound is a commonly used imaging modality for breast cancer diagnosis and prognosis but suffers from false positives, false negatives and interobserver variability. Deep learning (DL), a subset of artificial intelligence, has the potential to improve the efficiency and accuracy of breast ultrasound. This article provides a comprehensive overview of DL applications for breast cancer diagnosis and treatment in ultrasound, including methodological descriptions of various DL models, and clinical applications on noise reduction, lesion localization, risk assessment, diagnosis, response evaluation and outcome prediction. Furthermore, the review highlights the importance of interpretability and small sample size learning of DL-based systems in clinical practice; specific recommendations for further expanding the clinical impact of DL-based systems are also provided.