Deep learning has shown great potential in healthcare by enabling end-to-end learning from complex, high-dimensional, and heterogeneous biomedical data. This review discusses recent applications of deep learning in healthcare, highlighting its opportunities and challenges. Deep learning approaches can help translate big biomedical data into improved human health, but there are limitations in terms of interpretability and ease of understanding for domain experts and citizen scientists. The review emphasizes the need for interpretable architectures that bridge deep learning models and human interpretability. Deep learning is particularly useful in clinical imaging, electronic health records (EHRs), genomics, and wearable devices. In clinical imaging, deep learning has been applied to predict Alzheimer's disease, segment cartilage, and detect diabetic retinopathy. In EHRs, deep learning models have been used to predict diseases, classify skin cancer, and predict future medical outcomes. In genomics, deep learning has been used to predict DNA- and RNA-binding protein specificities and to model chromatin marks. In wearable devices, deep learning has been used to predict energy expenditure and detect freezing of gait in Parkinson's disease. Despite its potential, deep learning faces several challenges in healthcare, including data volume, data quality, temporality, domain complexity, and interpretability. These challenges require the development of improved methods and tools that enable deep learning to interface with healthcare information workflows and clinical decision support. The review also discusses the importance of incorporating expert knowledge, temporal modeling, and interpretable modeling in deep learning applications for healthcare. The review concludes that deep learning can open the way toward the next generation of predictive healthcare systems that can scale to include billions of patient records and use a single, holistic patient representation to effectively support clinicians in their daily activities. Deep learning can also serve as a guiding principle to organize both hypothesis-driven research and exploratory investigation in clinical domains. The review highlights the need for further research in feature enrichment, federated inference, model privacy, and the integration of diverse data sources in deep learning applications for healthcare.Deep learning has shown great potential in healthcare by enabling end-to-end learning from complex, high-dimensional, and heterogeneous biomedical data. This review discusses recent applications of deep learning in healthcare, highlighting its opportunities and challenges. Deep learning approaches can help translate big biomedical data into improved human health, but there are limitations in terms of interpretability and ease of understanding for domain experts and citizen scientists. The review emphasizes the need for interpretable architectures that bridge deep learning models and human interpretability. Deep learning is particularly useful in clinical imaging, electronic health records (EHRs), genomics, and wearable devices. In clinical imaging, deep learning has been applied to predict Alzheimer's disease, segment cartilage, and detect diabetic retinopathy. In EHRs, deep learning models have been used to predict diseases, classify skin cancer, and predict future medical outcomes. In genomics, deep learning has been used to predict DNA- and RNA-binding protein specificities and to model chromatin marks. In wearable devices, deep learning has been used to predict energy expenditure and detect freezing of gait in Parkinson's disease. Despite its potential, deep learning faces several challenges in healthcare, including data volume, data quality, temporality, domain complexity, and interpretability. These challenges require the development of improved methods and tools that enable deep learning to interface with healthcare information workflows and clinical decision support. The review also discusses the importance of incorporating expert knowledge, temporal modeling, and interpretable modeling in deep learning applications for healthcare. The review concludes that deep learning can open the way toward the next generation of predictive healthcare systems that can scale to include billions of patient records and use a single, holistic patient representation to effectively support clinicians in their daily activities. Deep learning can also serve as a guiding principle to organize both hypothesis-driven research and exploratory investigation in clinical domains. The review highlights the need for further research in feature enrichment, federated inference, model privacy, and the integration of diverse data sources in deep learning applications for healthcare.