Developments in data science and data management technology have enabled myriad innovations in healthcare from sequencing the humane genome to delivering personalized care to patients. At least in theory.
According to Beck Olson, Computational and Data Science Researcher at the University of California, San Francisco (UCSF) Center for Intelligent Imaging, while the research, results, and science are sound, there’s been a significant delay in putting theory into practice when it comes to integrating data-driven insights into clinical workflows. “There are three obstacles that have, until recently, stood in the way of taking promising research from the lab to the patient’s bedside,” Beck shared. “First, the size of data sets is enormous and many organizations lack the ability to move data quickly, efficiently, and cost effectively. Next, the deep learning technology that is needed to analyze these rich data sets and provide the building blocks for actionable insights has also been out of reach for most healthcare organizations. And finally, the rate at which clinicians add more data to be integrated into existing data sets or used to create new data sets makes this an overwhelmingly complex task.”
However, the UCSF Center for Intelligent Imaging directed by Dr. Sharmila Majumdar and Dr. Christopher Hess has been working with industry partners like NVIDIA and Kheiron Medical to bring university research into the clinical realm to deliver actionable AI insights directly to EMR and PACS applications. “While it’s been great for researchers to showcase the success they’ve had in the lab, the ball is typically dropped when it comes to moving these insights into the clinical workflow,” he shared. “The framework we are developing incorporates new technology for data acquisition, inference, review and clinical practice, creating a training loop for the model based on clinical validation and feedback.”
Beck and the team at UCSF is working to bridge the divide through a unique collaborative relationship between researchers and clinicians at the university. “Clinicians know the pain points that data can solve and the researchers are able to figure out the technology to support them best, and avoid the ball being dropped on the way from the lab to the patient’s bedside.” Beck noted. “The last thing a clinician needs is another application, or another click between her and the patient. Clinicians are already experiencing significant burnout from trying to maintain electronic health records (EHRs) and other healthcare tools that were supposed to make it easier to focus on the patient. So we’ve really focused on ensuring that our models are robust and accurate, that data flows quickly and seamlessly between applications, and that data is put into the appropriate clinical application the first time.”
To this end the UCSF team have focused on building a robust framework in radiology with the end goal of integrating results with the clinical PACS and/or EHR, and delivering patient-centric data directly from the radiologist to the clinician. “Being able to take data rich images, feed them into a model, train the model, and iterate with constant feedback at scale and speed enable us to deliver vital information for patient care to the bedside. This wouldn’t be possible without a state-of-the-art data management infrastructure,” Beck explained. Being able to manage data isn’t just about being able to process terabytes of data in minutes, but it’s about being able to orchestrate all the data you need access to in near real time. “One of the biggest hurdles in AI for hospitals that lack the proper infrastructure are multiple silos of clinical data. This makes it hard to gain access to all that valuable data to train AI algorithms,” added Tony Turner, Strategic Partner Manager for Healthcare at NetApp.
In the end, the UCSF Center for Intelligent Imaging wants clinicians and researchers to remember one thing: the key to success is found in collaboration. “This is not just collaboration between clinicians and researchers, but also between clinicians and AI,” he concluded. “The models researchers create aren’t there to replace the radiologist, but to enhance what she can do. The radiologist has to review the output before incorporating the results into patient care, but they’ll be able to do so more quickly because the burden from routine work has been alleviated by AI and finally, as we are demonstrating with our work at UCSF, integrated into clinical workflows automatically.”
Learn more about data-driven healthcare here.
