When we think about big data and the impact that it has on the world around us, one field that everyone can relate to in accepting its importance is health care. Whether we like it or not, at some point we will all seek the help or guidance of a medical professional. While AI technology can never replace the important role each member of a healthcare team plays – be it a nurse, technician, or doctor – we can say that if it’s implemented correctly, AI solutions can help inform the staff’s decisions. Anyone that’s ever worked on a floor in a hospital knows one word crosses all departmental boundaries and experiential borders: triage. Triage, of course, refers to the order in which patients should be addressed or tasks should be completed. For me sitting behind a computer, triaging which article to work on next or which program to clean up first is primarily tied to meeting deadlines. In the health care setting, it’s a much more complex determination with a much greater impact.
So, where does AI come in? The applications of AI in health care are essentially boundless. It could be in using x-rays or 3D images to train a system to diagnose a disease in its early stages or, as stated earlier, in aiding health care staff in triaging and gaining insights into treating their patients.
Privacy-Preserving AI Solutions
Now that I’ve got you on board with the power that AI solutions can bring to modern health care, herein lies the challenge. As you probably know by this point, to implement anything in the field of AI and machine learning, these complex systems need to be trained using massive amounts of example data. We’ve all had fun filling out lengthy medical forms that essentially categorize us by every variable a data scientist could think up, right? So, aren’t there already seemingly unlimited training sets of data available? The answer to that is yes, there are, but sharing data is not that simple. HIPAA (Health Insurance Portability and Accountability Act of 1996) ensures that the data you provide to medical centers remains confidential so this information has to go through some sort of anonymization process to avoid the risk of exposing sensitive data. As important as confidentiality is for patients, it also makes it very challenging to build large training data sets because cleaning the data to remove privacy violations is an extensive job that needs to be done in-house before the data can be anonymized and shared with other health care providers and researchers.
This is where federated learning and related AI techniques come into play. Essentially, these techniques allow each “base station” of data to remain exactly where it is, unshared with any other group contributing to the overall data pool. The AI now has access to each individual base station of data in real-time while maintaining the strict privacy laws of HIPAA. This allows for greatly enhanced training data sets so smaller health care providers can work as a larger team, each drawing benefits from the collective learning taking place. The take away is that, often, the most challenging problems we face can be overcome if we are willing to work collectively. With federated learning and enhanced AI privacy, the possibilities are endless.