Highlights from the Health AI Summit
The Nordic.AI Health AI Summit was held on September 13th at the beautiful Ny Carlsberg Glyptotek in Copenhagen, Denmark. The summit gathered the region’s best thought leaders, technologists, researchers, and practitioners to discuss some of the biggest questions in modern health. Our mission for the day was to help empower modern, technology-enabled healthcare to better serve our needs as a global community.
Irakli Beridze, the Head of the Centre for Artificial Intelligence and Robotics at the United Nations, opened the day with a talk about how AI is being used to solve societal problems. He noted that while we are richer, smarter, and more connected than ever, AI marks a true paradigm shift, and has only recently been embraced by the UN. In fact, AI was only first discussed at the UN general assembly in 2015, and acknowledged as a “game changer” only last year.
When he founded the UNICRI Centre on AI and Robotics in the Hague, the goal behind the center was to progress the discussion on robotics and artificial intelligence governance to key stakeholders and policy makers.
The UN sees AI as a contributing factor to solving development goals including ending poverty and hunger, reaching gender equality, and towards climate action.
However, he noted that the major risks in AI are in often due to the pace of development. There are no international treaties, resolutions, or the like on AI at all globally. A cohesive response to AI is necessary so that no single actor can have outsized control and influence over of this domain. The main takeaway from this talk was that a stakeholder cooperation is absolutely critical on AI policy development. However, it seems that everyone wants AI for good, so agreements should be within reach.
Next Holly Krelle, a senior analytics specialist at the National Health Service (NHS), discussed the future of AI within the British healthcare system.
While the NHS has the 4th biggest workforce in the world, demands for medical care have risen, and the NHS is actually understaffed. While AI could help alleviate some of these demand issues, the NHS has had a difficult history with IT projects, including a scandal where where an a new NHS IT system implementation was aborted, costing over £10bn.
However, AI projects move forward, including the NHS App Library, and GP at Hand, the first digital-first primary care technology.
The NHS is actively doing research to monitor their technologies and their system impact, to see how these technologies change how healthcare works.
Jussi Paanenen, the CTO of Blueprint Genetics, discussed how using AI approaches for genome sequencing can help identify rare diseases.
Over 350 million people are affected by rare diseases, which can be be caused by a single disease causing variant within their DNA. But DNA contains over 5 million data points, so finding these variants is a daunting task. The number of sequences of genomes doubles every year, and by 2020 there will be over 30 million genomes sequenced. But who will help interpret these genomes? This is an opportunity that is ripe for AI.
The difficulties in applying AI in this area are the training data needed, which is both expensive and rare, and a deep understanding of both genetics and patient care, which requires a highly cross-functional team.
Jussi urged the crowd that while AI is hyped, it is important to be both critical and ethical, to avoid pitfalls like algorithmic biases.
Dr. Egge van der Poel, the Big Data Doctor, gave a high-energy talk about how big data is creating inequality, but that this can actually be a good thing. He began with the adage that “most people are other people”, and that studies made on small subgroups still help us learn about whole populations. What’s noise for one group can be a signal for another, and discoveries often happen through collaboration.
He urged working together across disciplines, sharing data, and having a common goal.
Kjell Arne Furnes, the CEO of Ably, talked about how hospitals can monitor patients in the future through smarter hospital beds. After learning that smarter patient monitoring can help relieve nurses of some of the burdens of heir work, Ably set out to create smarter hospital beds that collect the pulse, respiration, weight, sound, and temperature of hospital patients. These basic signals let nurses know when irregular events occur, without getting into tricky diagnostic processes. Kjell also advocated for the democratization of data.
Andreas Mattson, VP of Engineering at Kry, spoke about how technology is democratizing access to doctors. He pointed out that with healthcare costs increasing, it is important to take a proactive rather than a reactive approach to healthcare, noting that “We check up on our care more often then we check up on ourselves, isn’t that peculiar?”.
Through their service that makes online consultations with doctors or psychologists, Kry has now helped more than 400,000 patients, covering some 3% of all consultations in Sweden.
Jón Skírnir Ágústsson, research manager at Nox Medical , talked about how machine learning (ML) is being used to drive new research in sleep studies. Sleep studies have historically been analyzed manually by technicians, and the field is still far behind when it comes to adapting new technology. Sleep studies use a huge amount of data, needing around 1GB just to detect sleep apnea in a patient. Nox not only makes medical devices and sensors that can be easily distributed, they can analyze the data using ML, drastically reducing the manual labor done by technicians.
Mattias Paulsson, CEO of Lytics, talked about how AI is bring brought into dialysis care. Using a data driven approach, Lytics is able to predict hospitalizations of dialysis patients within a 30 day window and help detect patient irregularities and optimize drug dosage in patients, which gives treatment centers valuable information to help them give patients the right care at the right time.
Nenad Tomasev of Deepmind gave an overview of the great successes in using ML techniques on real world problems. He proposed a mindset for researchers looking to do the same: start with a clinical approach, and ask questions and gather data first from a clinical perspective. He noted that fancy ML models are not always necessary, and the main focus for ML researchers in this field should be on usability.
He also discussed Deepmind’s work on 3D eye scans, and noted that the 2- stage architecture approach that was used was beneficial for interpretability.
Jonas Tyle Petersen of the DTU Digital Innovation Hub discussed the new collaboration of the lab and Nordic.AI to help promote the growth of new AI health startups in the region. The program includes access to a healthtech supercomputer, mentoring, and hackathons that generate connection between practitioners and startups to solve real-world problems.
Lars Maaløe, CTO of Corti, talked about how Corti provides decision support to emergency call takers in real-time. He demonstrated how ML can help improve the diagnostic accuracy and pace of decision making over emergency calls, and how Corti has built hardware to solve the issue of integrating this type of decision making software into emergency call sectors. He also underlined that it is important to let medical professionals know when there is uncertainty within a prediction, and to understand that decision support does not necessarily mean always taking the decision made by an AI.
Sune Askjær, Principal Scientist at LeapBeyond, talked about how AI is making headway in drug discovery. Drug discovery is a trial-and-error process, and the average cost for new drugs to enter the market clocks in at close to $3 billion. Learning algorithms can function as “experts” of a sort, and can learn from historical data to help speed up this process. The space of drug-like compounds is 55 orders of magnitude larger than what has already been attempted, and it is important to use computers to help explore that space.
Fred Abrahams, Associate Program Director at Human Rights Watch, talked about how AI has become a hot topic for human rights research, for both good and bad. When used for good, this technology can for example help discover and combat human rights violations. Recently, drone images of violence against the Rohingya in Burma were critical evidence used to tell the story of the atrocities being committed. However, this technology can also endanger human rights, and issues such as privacy, bias, freedom of expression, security, surveillance, and responsibility must be grappled with. For example, who is responsible for the war crime if a robot kills a civilian? The Partnership on AI is one initiative seeking to understand these issues.
Nicola Rieke, a Senior Deep Learning Solutions Architect at Nvidia, spoke about how computational imaging is everywhere in healthcare. Deep learning is already part of this in practice, such as in GE scanning procedures. The next generation of healthcare will need to include AI in everything from the devices used to how health information is stored and disseminated. She gave an overview of major research trends in medical imaging, including uncertainty estimation, network innovations, unifying heterogeneous data, and weakly supervised learning.
The day closed with a panel ethical issues within AI and healthcare with Irakli Beridze, Holly Krelle, and Fred Abrahams, moderated by Christian Sejersen from the LEO Innovation Lab.
The panelists agreed it was important to make sure people in less developed areas do not become guinea pigs for our technologies, and ensure the same types of standards apply across different types of communities. They especially urged towards informed consent for when technology is used.
Responsibility issues in automated healthcare shouldn’t themselves be automates from humans to machines — there must be a human-in-the-loop.
That’s a wrap for the Nordic.AI Health.AI Summit! Sign up for mailing list or follow us on Facebook or Twitter to learn about future events.
All photos courtesy of Jonas Clark Jeppesen, creative editor of Tame.
