Machine Learning on Health Data

Health systems routinely capture vast amounts of data of many different modalities; recently, the amount and types of data have dramatically increased.  The data science revolution is now pushing the boundaries of artificial intelligence research. Advances in machine learning have opened up new applications leveraging an increasing number of data types. 

Machine learning on health data is an important focus area of the department. Many of the faculty are engaged in research that brings together Artificial Intelligence, Machine Learning and analysis of health data.  This research both includes developing new machine learning techniques specifically tailored for health data as well as utilizing machine learning to obtain insights from health data that can improve patient care.

One example project is a collaboration with the Anesthesiology Department at UCLA.  Surgeries are remarkably safe, yet there are often complications. These range from rare complications such as mortality to fairly common ones such as acute kidney injury, which affects one in six patients.  If we can predict which complications are likely for which patients, we can prepare for them, and make extra efforts to avoid them. Computational Medicine, in collaboration with UCLA Anesthesiology, developed a machine learning system to predict surgery complications based on the health records of a patient before surgery. Our system can find clues in a patient's history that can increase the risk of complications. Since there are more than 50,000 surgeries each year, our system will positively impact the care of many patients.

Other examples of machine learning on health data projects in the department include predicting retina related functions from eye images, and predicting depression from sensor data. The department’s projects are always in close collaboration with clinicians with the goal of directly improving health care.