Paging doctor data: Machine learning and the future of healthcare by Allie Nicodemo August 17, 2017 Share Facebook LinkedIn Twitter “I’m excited about realizing data-driven patient care and healthcare policies,” said Byron Wallace, assistant professor of computer and information science at Northeastern. “I think machine learning will be key in getting us there.” Photo by Adam Glanzman/Northeastern University Healthcare has long been at the forefront of public and political debate. While policy progress appears to be gridlocked, advancements in technology blaze ahead. One area scientists and physicians both consider ripe with potential is machine learning, an emerging field that is closely linked to artificial intelligence and Big Data. What robots, supercomputers, and algorithms lack in bedside manner, they make up for in the ability to sift through mountains of medical information. “Machine learning is the art and science of uncovering structure in data,” said Byron Wallace, assistant professor of computer and information science at Northeastern. Wallace is organizing the Machine Learning for Healthcare Conference, which will be held at Northeastern on Friday and Saturday. The conference convenes two camps of experts who don’t traditionally collaborate—computer scientists and medical researchers. Among the topics up for discussion is a Food and Drug Administration-approved product called Arterys that uses deep learning and cloud computing to help radiologists read MRIs with improved speed and accuracy. Another researcher will explain how the same types of algorithms Amazon and Google use to predict your shopping preferences can be leveraged to predict diseases. “I’m excited about realizing data-driven patient care and healthcare policies,” Wallace said. “I think machine learning will be key in getting us there.” Here, Wallace explains more about why machine learning is just what the doctor ordered. Why is machine learning being explored in the context of healthcare? The rapid proliferation of healthcare data—from electronic health records, to published literature, to relevant social media content—provides an unprecedented opportunity to improve and personalize patient care using data-driven, empirical approaches. But doing so requires scalable approaches to processing and making sense of this torrential volume of data. Broadly, machine learning methods provide a possible means of realizing this aim. What is the potential for machine learning to transform the healthcare field? Machine learning approaches have the potential to help providers and individuals navigate large volumes of data to improve health decisions and outcomes. For example, machine learning may automatically identify patients at pronounced risk for a particular adverse event, or it may be used to identify “sub-groups” of patients that may be clinically meaningful. There is also the potential for machine learning to aid in understanding broader, population-level trends. This has given rise to a subfield of “computational epidemiology,” which uses machine learning and other computational models to study public health. What do you think are the biggest challenges facing healthcare? How does machine learning fit in? One major challenge is the sheer volume of health-related data that is now available. Much of this is “unstructured,” meaning it is not immediately amenable to processing and synthesis. Furthermore, healthcare data originates from a number of heterogeneous sources, which poses another set of challenges. Machine learning approaches constitute a toolset that can potentially consume large-scale, heterogeneous, unstructured data and produce actionable outputs that might inform decision-making and patient care.