No book on healthcare analytics would be complete without a discussion of deep learning. In recent years, unparalleled results have been achieved in the areas of speech recognition, face recognition, language understanding, and object identification using deep learning algorithms (Goodfellow et al., 2016). Almost in tandem have been the breakthroughs in healthcare; from cancer detection on pathology slides and radiology scans to predicting mortality and readmissions, time and time again, results are being seen that rival (and in some cases surpass) committees of expert physicians.

We have already discussed some aspects of deep learning briefly. In Chapter 1, Introduction to Healthcare Analytics, we discussed a historical paper that is seen by many to have been a seminal event for deep learning as a field, in Chapter 3, Machine Learning Foundations, we discussed deep neural networks as a medical decision-making framework, and in Chapter 8, Healthcare Predictive Models – A Review, we discussed some studies that used deep learning algorithms to achieve their results. Let's now briefly discuss what deep learning is and what distinguishes it from traditional machine learning and neural networks. We will then discuss some promising studies that use the following subtypes of deep learning algorithms in reaching their conclusions:

• Deep feed-forward networks
• Convolutional neural networks (CNN)
• Recurrent neural networks (RNN)

I have decided to keep discussion of the theory of deep learning out of this book since a brief, simplistic chapter on deep learning theory would not do the field justice. There are, however, some excellent courses on Coursera (www.coursera.org) that explain what deep learning is mathematically in a very easy-to-follow manner.