Deep learning for biomedical applications

The types of data most commonly used for machine learning in biomedical research, including electronic health records, imaging, -omics, sensor data, and medical text, are complex, heterogeneous, poorly annotated, and generally unstructured. However, gaining knowledge and actionable insights from all these data sources is a key challenge in implementing personalized medicine and next-generation healthcare. Deep learning, which describes a class of machine learning algorithms based on neural networks, provides effective opportunities to model, represent, and learn from such complex and heterogeneous sources. Here, we review how this paradigm has already affected healthcare, we note limitations and needs for improved methods and applications, and we discuss the challenges to implement and deploy augmented human intelligence based on deep learning in the clinical domain.