Technology has transformed our lives as consumers, but it’s only just starting to make a difference to our health: you can look up your symptoms online, track your heart rate with a smart watch and expect your doctor to look up your previous prescriptions. In the future though, technology might give us precision medicine personalised to our own immune systems and genetics. Before that, it can make healthcare more efficient – giving doctors more time to look after patients.
As AI and machine learning become more important in medicine, Microsoft is taking existing investments in its healthcare data platform, cloud-scale databases and AI tools and using that as the basis of Healthcare NExT – a set of partnerships to develop solutions that have the right tools for healthcare, at a scale that can help treat and protect everyone. It’s based on the MSR NExT initiative for turning research into new lines of business that created new Microsoft products like the Bot Framework and Azure Sphere; in 2015, Microsoft CEO Satya Nadella decided to take the same approach to healthcare and the first pieces are now starting to emerge.
It all starts with Azure, head of Healthcare NExT Peter Lee explained to us, because “healthcare is becoming more data intense and so it’s increasingly dependent on the kind of hyperscale infrastructure in our cloud. In the future of precision medicine with highly personalised diagnoses and precisely targeted treatment, your genetics become really fundamental and a single human genome takes 100 to 200 Gb of storage and several hours of compute in a large cloud VM.”
A cancer clinic or a neonatal intensive care unit wouldn’t have the computing power in-house to handle that, let alone the details about your immune system, medical imaging, information from wearable sensors and even information about your social background that will need to be combined with it – because those signals are very specific to you and they also change. All areas of medicine are going to have massive compute and storage workloads, whether that’s handling precise imaging for segmenting tumours to assist radiologists and oncologists, tertiary genomic analyses or the massive task of decoding the human immune system.
That’s something Microsoft is working on with Adaptive Biotechnologies, in the hope of creating a universal blood test that could provide early diagnosis of cancer and other diseases. “The immune system that fights disease has T cells and B cells,” Lee explains. “There’s a receptor on the surface of the T cell that has a snippet of DNA that genetically programs the cell; that DNA sequence determines what pathogen the T cell can bind to, and when that binding occurs the T cell is programmed to destroy that cell.”
When a T cell encounters a pathogen that it matches, not only does it destroy the cell – it also clones itself so you get more T cells to deal with the same pathogen. A test that would take four to six hours could extract the DNA sequence for the million T cells found in a typical blood sample; knowing what T cells are common in your blood would tell a doctor what your body is trying to fight. “It’s a big machine learning problem and it has structural similarities to the translation from one language to another, only it’s the translation of T cell receptor sequences to the language of antigens,” Lee explains.
Translating languages is something machine learning researchers have made huge strides with using deep learning; Lee says medical researchers are more and more confident that with the right training data they can build machine learning models to detect what your T cells are battling. “If this works out, you could take a universal blood test every year that tells you what’s in your body. It’s like a Star Trek tricorder but you need a needle!”
Even if a universal test isn’t possible, blood tests for the early detection of ovarian cancer – something we don’t have an early test for today – herpes, type one diabetes and coeliac disease look achievable, Lee says. “There’s no test for coeliac disease that doesn’t damage your body; this could be a definitive and benign test.”
Why is a technology company like Microsoft getting involved in such fundamental medical research? “Because if it’s ever to become reality it requires the computing scale and machine learning capabilities that only Microsoft and a very few other companies have,” Lee says.
Initiatives like this may not bear fruit for years or even decades; others are ready for researchers to work with now.
Microsoft already has a certified, HIPAA-compliant commercial cloud service for genomics analytics, and it’s working with institutions like St Jude Children’s Research Hospital and the Integrative Brain Research Institute to build clinical genomics databases in the cloud for childhood cancer and infant mortality that researchers around the world can access. The Broad Institute worked with Microsoft to create the open source Cromwell project to orchestrate genetic analytics using Azure Batch; Answer ALS is using Cromwell to compare the genetics, epigenetics, RNA, proteins and cellular metabolism of patients with ALS to that of healthy people to understand more about the disease.
Another area that’s divided between tools that are ready to work with today and longer-term efforts is what Lee calls ‘empowering people on the front lines of healthcare’ using natural language processing, speech recognition that can understand multiple people in the same conversation and machine reading. For medical researchers and even healthcare administrators, machine learning can help with mapping schema so that data from different healthcare organisations can be used together. Bots will be able to automate routine tasks so doctors and nurses can spend more time with patients – and when they are with patients, they’ll be talking to them and listening to them, not staring at the screen and poking at the keyboard.
For every hour doctors spend treating patients, they spend two hours on paperwork – looking up patient records or filling them in. EmpowerMD, the system Microsoft is building with Nuance could change that by observing the conversations doctors have with patients and then learning to automate all the documentation, as well as letting the doctor look up an old record or ‘write’ a prescription with a voice command. It could even suggest a diagnosis to match your symptoms or spot when you’re asking about a new medication and check your medical history to see if you’re likely to be allergic.
That’s just one of the partnerships Microsoft has formed in key areas; it’s working with Providence, one of the largest US health systems to create a ‘clinic of the future’ for cancer care and with Boots and Walgreens to create a cloud AI platform to integrate information between your doctor, the pharmacy and any other health services you use. It’s working with Novartis to use AI to help pharmaceutical researchers to mine large datasets to make it faster and cheaper to develop new drugs, and with Humana to build a system that can create personal treatment plans and drug schedules that are easier for patients to follow. Instead of custom solutions, the idea is to create systems that can scale for any health provider who needs the same kind of system.
Some of the projects are much smaller. A hackathon at Microsoft Research in Israel created a bot that tries to match patients with clinical trials using a sequence of questions based on the trials available: if any pharmaceutical companies pick that up it could make it far easier for patients with unusual symptoms to find help, and make sure new drugs are tested thoroughly. None of this technology is going to replace doctors, but it could take a lot of the tedium and inefficiency out of healthcare, even without new breakthroughs.
This is the second of three posts about Microsoft’s innovations in Healthcare.
Read the first in the series, “What Microsoft offers the healthcare market“.
Grey Matter has a Cloud Services team who are experts in managing and enabling Microsoft cloud technology and Azure-related projects. If you need their technical advice or wish to discuss Azure, O365, Visual Studio, or your use case, contact them on +44 (0)1364 654200.