Hospitals have vast amounts of data providing huge potential for the medical sector to utilize AI, however, there are challenges in aggregating the data, munging the data, and understanding the data. 'We're actually at a very interesting time in healthcare where AI/Deep Learning typically isn't the limiting factor -- there are amazingly powerful tools at our disposal. The primary challenges faced are accessing the data in all the various forms it may exist within an institution, transforming the data which was collected primarily for clinical purposes into a form amenable to machine learning, and being able to bring together the data scientists and the clinicians to make sure that we don't lose our way to just optimizing AUCs (or other standard error metrics)-- but focus on creating insights that we can bring back to the bedside.'
At the Children's Hospital of Los Angeles, the team are using deep learning in the Pediatric Intensive Care Unit (PICU) to predict individual physiologically acceptable states at discharge from a PICU. David Ledbetter, Senior Data Scientist at the hospital explains that they deal with PICU for two main reasons:
David and his team focus on combining the PICU data, state-of-the-art algorithms (primarily Recurrent Neural Networks (RNNs)), and the clinical expertise of our CHLA critical care physicians to generate data products we can bring back to the bedside to help the clinical team provide the best treatment possible.At the Deep Learning in Healthcare Summit in Boston this May 24 - 25, David will be presenting his most recent work in the space. Starting out his career in remote sensing (radar, satellite, sonar, etc.), David was fortunate to learn about the fundamentals of detection theory (how to find a signal in noise) before exposure to machine learning and deep learning. 'It provides a great frame of reference in which all of the advances in AI are easily described. I got an opportunity to perform some analysis on the CHLA PICU data, and spend some time in the CHLA PICU itself and after that, there was no turning back.'
Every ICU encounter contains a wealth of information about disease progressions, treatments, and patient outcomes. AI enables us to explore the information available from ICU encounters and generate models that understand complex relationships between drugs, interventions, and patient well-being. In effect, we use AI to learn from those experiences and apply what’s been learned to provide the best care possible for each child that comes through our doors.Positive social outcomes are something the industry is focusing on intently as of late, and AI provides exactly this with the potential to aid drug discovery, speed up diagnoses, help with medical imaging, and all sorts of areas in medicine and health. All healthcare data provides experiences to be gleaned, and 'AI provides us with an opportunity (and a framework) to ensure that we are learning as much as possible from those hard-learned lessons. With that accumulated knowledge in hand we’ll continue to improve our ability to predict outcomes or diagnoses, but more importantly, better understand how (and when) to intervene in order to improve outcomes for patients across the entire industry.'