Image by Gerd Altmann from Pixabay
Artificial intelligence is already widely entrenched in healthcare industry, so the question should not be, "Will AI happen or not?" said David Houlding, CISSP, CIPP, principal healthcare lead, at Microsoft, in an interview with MD+DI. “It's very much 'What tasks will AI help with, and what will it do and what will it not do?’ ” he continued. “I think we're still learning some of that, but it's largely a proven technology, not just for healthcare providers but for payers, for pharmaceuticals, for life sciences—really every segment of healthcare is using AI some way, shape, or form.”
Houlding will speak as a panelist in the December 4 BIOMEDevice San Jose panel discussion, “The Growing Artificial Intelligence Application Across Medical Devices.”
Houlding told MD+DI that what AI and machine learning are really good at is pattern detection. “The idea there is for AI to help healthcare professionals improve their decision-making and make sure they are not missing any patterns or correlations.” In this way, he explained that AI can be used as a tool to augment, assist, and empower healthcare professionals with performing tasks such as segmenting images, among others.
AI can also be used to help detect rare conditions that a GP might never see and therefore not know what to look for, as well as to triage patients whose scans could indicate an imminent critical condition. “There is also a way AI can support remote patient care solutions, by processing data near real-time from medical devices monitoring patients, and providing insights that assist healthcare professionals in proactively detecting if the patient’s health is trending in the wrong direction. Then the [doctor] can intervene to course-correct and avoid episodes that can not only be costly, but decrease the patient’s quality of life,” explained Houlding. “So it's about improving patient care, reducing costs, and improving patient engagement and experiences as well.”
He cautioned that although AI has fantastic potential, those in the healthcare industry need to learn how to use it well. “AI is very task-oriented—it’s like pattern matching, so it’s not like AI can replace a job. It’s more like AI can help with the specific tasks,” Houlding said, noting that this is a key reason why there will have to be a human in the loop.
“Sometimes AI gets it wrong,” he said. “AI can be biased, so one of the challenges is getting enough high-quality data to train models.” Houlding explained that if you build a model with either not enough data or data that is not high quality, what you end up with is something suboptimal that will produce insights with a higher error rate and that could be biased.
He noted that a lot of data models that AI is based on are being built from information from single organizations, which is not ideal. “If you're drawing training data out of just one silo, you've only got a limited quantity of data, and potentially also limited quality, so we've got to figure out a way in the industry to enable building models from data from across organizations, and enabling filtering of data based on detailed provenance information to ensure only the highest quality training data, and highest quality models trained from it.”
One solution Houlding envisions is to have a whole consortium of health systems that are all using a given type of medical device, and all are getting data from the use of this type of medical device. “If they can share the training data with provenance information, and models, as well as results of their testing and the validation of those results, then as an industry we can advance AI significantly, and learn to trust new AI models much faster,” he explained. “Blockchain has a key role to play in enabling this kind of collaboration across a consortium of health systems using medical devices of a given type.”
Another challenge for using AI is in the way medical devices are regulated. AI and machine learning can be continuously learning and changing over time. Houlding asked, “How do you regulate that to ensure the medical device with continuously learning AI complies with regulations on an ongoing basis? How do you know that every incremental Improvement in the AI system that's continuously learning still works, as well [as] or better than the original medical device with the baseline AI model it came with?”
Houlding hopes that attendees to the session will take away a better understanding of these opportunities and challenges and the ways in which AI can successfully be used in the medical device industry to improve patient outcomes, reduce costs, and improve patient engagement and experiences. He said that anyone involved in manufacturing or using medical devices or solutions that connect with medical devices should attend. “It’s about AI and medical devices both being a key part of the solution, and increasingly this AI is cloud based. Anyone involved in medical devices would find the session useful.”
The panel discussion, “The Growing Artificial Intelligence Application Across Medical Devices,” will be held on Wednesday, December 4, from 1:45 to 2:30 p.m., in Booth #641, at BIOMEDevice San Jose. It will be moderated by Srihari Yamanoor, President at DesignAbly, with panelists Houlding; Dr. Manikanda Arunachalum, senior vice president, corporate developments & investments & head of healthcare, founding member, board of directors at Beyond Limits, Alliance for AI in Healthcare (AAIH); and Bimba Rao, director of engineering, AI group at Siemens Healthineers.