Roxanne Nelson, RN, BSN
November 28, 2018
Software giant Microsoft has entered the healthcare arena, and improving cancer care and diagnostics is high on its list, company executives told Medscape Medical News during a recent visit to the Seattle headquarters.
One of the company's predictions: "Ten years from now, cancer will be a solved problem thanks to interdisciplinary, groundbreaking approaches that will enable researchers and clinicians to compute driver mechanisms of cancer, as well as to understand, detect, diagnose, and treat patients at an individual level.
"I believe executable biology will play a key role in tackling this enormous challenge," said Jasmin Fisher, PhD, principal researcher at Microsoft Research Cambridge, United Kingdom, in the programming principles and tools group.
Fisher, who is also an associate professor of systems biology in the Department of Biochemistry at the University of Cambridge, has focused her career on an innovative approach called executable biology — the design and analysis of executable computer algorithms describing biological phenomena — and in particular, cancer biology.
When the United States declared war on cancer in 1971, few could have imagined the sheer enormity of the task that lay ahead. Even fewer would have envisaged that in addition to traditional laboratories equipped with test tubes and microscopes, scientists and engineers would be using algorithms, coding, and computers to ramp up the intensity of cancer research.
The concept of using a multipronged approach to improve cancer care is not a new one, but a growing number of tech companies are testing the waters and approaching cancer research as a tech problem instead of a medical one. Although most research is still conducted in traditional settings, several of the large tech companies have stepped up to the task and are now delving into the realm of healthcare.
One of these is the software giant that transformed its dominance of the desktop and has become a cloud computing powerhouse — Microsoft.
"It's terrific that Microsoft is on this," commented Eric Topol, MD, executive vice-president, Scripps Research founder, and director and editor-in-chief at Medscape. "Their proficiency in AI [artificial intelligence] can help improve outcomes for cancer, and ultimately even promote its prevention.
"A systematic, multimodal, data-driven approach (genomics, imaging, pathology, drug responsiveness, and more) will be vital, and the parts to that are getting assembled," he added.
Internet Startup in an Incubator
Microsoft's foray into healthcare is relatively recent. In 2014, the company formed Microsoft Research NExT, a division of about 500 scientists using innovative technologies to develop commercial products.
Last year, Microsoft CEO Satya Nadella tasked Microsoft Research NExT with the job of forming a division devoted to improving healthcare. Cancer care was high on the list, and the span of possibilities seemed endless — everything from using machine learning and natural language processing to develop innovative treatments, to improving tumor scanning, to "moonshot efforts" focused on programming cells to fight diseases.
Peter Lee, PhD, corporate vice president, Microsoft Healthcare, explained that the goal of his group is to work on technologies for better and more efficient healthcare, with a special focus on AI and cloud computing.
The goal of Healthcare NExT is to transform healthcare, integrate research and health technology product development, and establish a new model at Microsoft for strategic health industry partnerships, he said. Thus far, Microsoft has entered into collaborations with a diverse group that includes the Fred Hutchinson Cancer Center, St. Jude Children's Research Hospital, Adaptive Biotechnology, the University of Pittsburgh Medical Center, and the British Columbia Cancer Agency in Canada.
For Microsoft Healthcare, which evolved from NExT, the new mandate was so broad that Lee likened it to being dropped in the middle of the ocean and told to find dry land.
"This is sort of an internal startup in an incubator," Lee told Medscape Medical News. "When this first happened, I almost felt victimized in the beginning, because this was a whole new way to go from research to product."
The NeXt model is driving a new strategy for Microsoft in healthcare. "But when you think about doing that, the first question you ask yourself is, 'What right do we have to exist in healthcare at all?' " he said.
It's not that uncommon for big tech companies to believe that they should have a piece of the healthcare pie, Lee continued. "And it's not uncommon for big companies to do that with maybe even some arrogance. So it's important to be grounded, and I would put it this way and ask the question, if Microsoft disappeared today, what potential would be lost for the healthcare world?"
The reason to ask that question is that it creates a discussion, he emphasized, "in that, what do we think the world of healthcare should be like if we want it to be better? And then we can see what role Microsoft can play in helping that along."
Lee explained that their investments are focused primarily in three areas: relevance, empowerment, and transformation.
Relevance is about making sure that the co-called cloud and all of the maturing AI are a standard utility for helping healthcare organizations do a better job. "That's where a lot of the focus on data interoperability has gone into," he said. "We've been trying to work as fast as possible, because right now, all of the healthcare providers and payers we work with are in the process of migrating to the cloud. If we can set up the cloud so there is secure and frictionless movement of healthcare data, we'll be in the right place."
He added that he would like all the clouds to be equally good at that, but "we would like to be there first, as we're competitive that way."
Empowerment is about putting the right tools in the hands of people in the front lines of healthcare delivery — physicians, nurses, patients, and administrators. "That means that all the great tools that we have for collaboration are compliant and ready for all health IT systems, but we're also trying to innovate and make new AI products," Lee said.
Transformation has to do with precision medicine, in which the focus has mostly been in oncology. Genomics, immunotherapy, and advanced imaging are becoming increasingly common in cancer care, and the information in those areas — the datasets — are much larger than can be comprehended by human beings, Lee noted.
For example, in cancer genomics, information has grown exponentially, with hundreds of research articles being published on a daily basis. "So now just imagine a tumor molecular board trying to evaluate all of this information — there is just no way they can possibly be up to date with that," he said. "So this is another area where machine learning AI can be so important in augmenting what people can do."
There are, of course, benefits for the company, Lee pointed out. These products can open new markets for Microsoft, and according to some analysts, healthcare cloud computing could turn into a $35 billion market by 2022.
Hoifung Poon, PhD, director of precision health natural language processing at Microsoft, agrees with Lee that progress in precision medicine has often been slow. Genome-scale knowledge and reasoning have become bottlenecks in the effort to sort through the complexity of cancer and other diseases.
"Medicine today is imprecise," he told Medscape Medical News. "There is a lot if opportunity and low-hanging fruit. AI, natural language processing, and machine reading are three areas that are pretty ripe for potential opportunities."
Poon's focus is on a data-driven strategy that uses a branch of AI called machine learning, which may be able to perform the legwork that is needed to clear up the bottlenecks and make precision medicine a reality.
For the top 20 prescription drugs in the United States, 80% of patients do not respond to treatment. Thus, up to one third of healthcare spending is wasted. Poon noted that it can take hours for a tumor board to review the "omics" data from one patient to make an appropriate treatment decision. With 1.7 million new cancer cases annually, this is clearly not scalable.
Hundreds of new cancer drugs are in development, and research is continuous. The "problem is there's too much to read and too many drug combinations, which make it difficult for doctors to choose the best regimen every time," he said.
Machine learning has tremendous potential to streamline information and dramatically speed up the process of choosing the most effective therapeutic regimen for each patient, Poon commented.
One of the most important applications will be to help researchers siphon through the medical literature, he continued. PubMed, for example, contains 27 million abstracts. Two new abstracts are added every minute, which amounts to over one million per year. Sorting through the literature manually is a very slow process, owing to the sheer volume, and it is made more complicated by inconsistencies in the language used by researchers.
"Using machine reading to extract knowledge from the oncology literature can increase curation throughput by 100 times," explained Poon. "This can empower the curator to go much faster."
To achieve that goal, Poon and his team are developing natural learning processing technology to convert text into structured databases. By automatically reading millions of biomedical articles, genome-scale knowledge bases will be created.
Another application will speed up the process of evaluating new therapies and genetic targets. Currently, tumor boards are limited to reviewing single genes and drugs, he explained. Next-generation molecular tumor boards will be able to factor in interactions among mutations and recommend combinations of drugs to attain synergistic effects and more effectively prevent relapses.
With hundreds of candidate targeted drugs, there are tens of thousands of possible combinations. Poon explained that they are developing a machine learning approach that models complex drug interactions and off-target effects. The ultimate goal is to assist clinicians and scientists with the basic research, and then use a Microsoft Azure cloud computing–based tool to enable physicians to model optimal treatments.
Poon's team is collaborating with the Knight Cancer Institute at Oregon Health and Science University, Portland, to develop machine learning methods that will integrate genomics knowledge with experimental data to help personalize drug combinations in acute myeloid leukemia.
InnerEye Already Moving to Clinical Setting
The goal of AI in medical care is to augment the skills of physicians and other providers. This concept is being targeted toward improving the productivity of oncologists, radiologists, and surgeons when working with radiologic images.
Project InnerEye is an innovative machine learning tool that can assist radiologists in identifying and analyzing 3D images of malignant tumors. Tarapov pointed out that currently, this needs to be done manually, and it is labor intensive and time consuming. "It can take hours for each patient, because the tumor and the healthy tissue need to be mapped out before beginning radiotherapy," he said in an interview. "The outlining is done in 3D and needs to be repeated for dozens of slices — wherever the tumor is visible."
Doing this manually is not only time consuming but can be inaccurate and expensive, he pointed out. Conversely, AI can be trained to recognize the tumor and healthy tissue and do this automatically. Using this technology, images are analyzed pixel by pixel to precisely determine the location of the tumor, after which the surrounding organs are delineated or contoured.
"AI has the capability of identifying the tumor and nearby structures in 3D and can do segment contouring," said Tarapov. "It can do this with 90% accuracy and dozens of times faster than doing it manually."
As an example, it takes 40 minutes to delineate the bladder in prostate cancer treatment, but with this tool, it takes less than a minute.
Unlike some projects at Microsoft Healthcare, which are still in their early stages of research, InnerEye is already moving into the clinical setting. Tarapov explained that they have both clinical and commercial partners. For example, at Addenbrooke's Hospital, in Cambridge, United Kingdom, InnerEye is being used to treat prostate cancer patients. Physicians can process patients faster, treatment is begun sooner, and radiotherapy is delivered with more precision, Tarapov said. Brain tumors are next on the list.
InnerEye is being evaluated in clinical trials. "We need to know how physicians are using it, and we are getting positive results," he said. "We are working together to build better models."
Commercial partners include TeraRecon, which is using InnerEye to create an "AI results viewer" for radiology, and Intuitive Surgical, which is working with the InnerEye team to explore machine learning applications to improve surgical outcomes.
Fueling Genomic Research
St. Jude Children's Research Hospital in Memphis, Tennessee, founded in 1962, is focused on children's catastrophic diseases, with an emphasis on leukemia and other types of cancer. Researchers at St. Jude's have been sequencing and deciphering childhood cancer genomes for nearly a decade. Sharing research and scientific discoveries, as well as collaboration, have been always been part of its mission.
However, downloading public data is a labor intensive and tedious task. It was becoming increasingly difficult to download and use data, owing to the volume that had been amassed. Thus, the researchers at St. Jude began to explore ways to facilitate data sharing with the global research community. They realized that they needed a platform in which collaboration and sharing could take place effectively and easily. This need ultimately led to a collaboration with Microsoft's genomics group.
"We are part of a larger team that has what we call a 'multi-omic' agenda, so we're not just looking at genomics but at things like metagenomics and immunogenomics," explained Geralyn Miller, director of Microsoft Genomics (AI and Research). "Microsoft has been doing work in genomics for over 10 years, and we were kind of at a pivot point."
She pointed out that genomics was coming into the mainstream primarily because of trends on the sequencing side. "It was getting more accurate, and so moving from a research focus into a mainstream clinical adoption," said Miller. "We looked at trying to take some of the core assets that Microsoft had and put research into the broader world — get this out of the lab and bring it to people who can use it, and St. Jude is a perfect example."
Members of the St. Jude team met with Microsoft, and from that meeting came the realization "that we are synergistic," said Miller. "We are a software company, and we bring that knowledge and expertise to the table. St. Jude had the knowledge and access to clinical information, real patients, and the research expertise. So put these two worlds together and we can do something meaningful."
The result was the launch of the St. Jude Cloud, for which data are stored on Microsoft Azure, an ever-expanding set of cloud services that can handle datasets on the massive scale required for large genomics studies. Another collaboration is with DNAnexus, a bioinformatics and genomic data management company that leverages Azure to provide an open, flexible, and secure cloud platform to support Microsoft Genomics services as well as other genomics analysis tools.
Currently, the platform contains 5000 whole-genome, 5000 whole-exome, and 1200 RNA-Seq datasets from more than 5000 pediatric cancer patients and survivors. By 2019, St. Jude plans to have 10,000 whole-genome sequences available on its cloud.
Miller emphasized that her group was "struck by some of the stories they told us about how hard it was for a small team to do some of the research in genomics because of fundamental stumbling blocks."
These included data sharing — to get sample sizes big enough to reach statistical significance, they needed to share data with researchers all over the world, but that was difficult. Another problem that computer processing took time. "We knew that we could help with both of those problems," she said.
The St. Jude Cloud is now the largest pediatric oncology dataset in the world. "What this has really done is lower the barrier for researchers to collaborate and to increase their productivity and allows them to shift their focus to the really high-value areas," said Miller.
Universal Screening Test
The promise of liquid biopsy, in which information is obtained from a blood sample instead of by invasive tissue biopsy, has garnered a great deal of interest among clinicians and patients alike. A simple blood test would be less stressful, cheaper, and more accessible than many current screening and diagnostic modalities.
Adaptive Biotechnologies, a Seattle-based biotech company, is interested in taking this concept a step further. It is pioneering the use of immunosequencing in patient care. In collaboration with Microsoft, its goal is to develop a universal screening test.
Jonathan Carlson, PhD, director of immunomics in the Healthcare NExT group at Microsoft, explained that theoretically, a simple blood draw would be used to screen for conditions that include infections, cancers, and autoimmune disorders in their earliest stage, when they can be most effectively diagnosed and treated.
"The single blood test is a vision, and people are looking at the liquid biopsy from different angles, such as cell-free DNA and different biomarkers," he said. "Our focus is very specifically on T cells, and the premise is that our immune system is the best diagnostic tool. If we can learn to decode what the immune system already knows, then we can have a universal diagnostic.
"The basis of this project is learning the T-cell Antigen Map," he said, "which is learning the mapping from a T-cell receptor sequence to the antigens it targets."
Technology developed by Adaptive can sequence the T-cell receptors, with the goal of mapping the genetics of the human immune system, or the immunome. A universal T-cell receptor/antigen map can be developed as a model of T-cell receptor sequences and the codes of the antigens they have been exposed to, he explained.
The immune system is designed to scan for and respond to the antigens that an individual comes into contact with. Advances in immune sequencing now allow T-cell receptors to be matched to hundreds of the antigens they recognize.
"We know that the T-cell receptors contain the information we need to diagnose disease, but we don't know how to decode it yet," said Carlson. "So the focus of our collaboration is to take the information they already have and decode it.
"Of course," he added, "the devil is in the details of how we are actually going to do that, but that's a high-level goal."
Adaptive and Microsoft have formed a partnership to train an AI tool to read that data and detect any diseases the immune system has recognized in the body. The initial focus will be on three classes of diseases: those that are often diagnosed in very late stages, such as pancreatic and ovarian cancer; autoimmune diseases that are typically hard to diagnose, such as multiple sclerosis; and infectious diseases that can remain in the body and reoccur, such as chronic Lyme disease.
Regarding the company's involvement, Carlson explained that Microsoft "is an example of a company that is sitting on the data and has the core technology, and it's an interesting application for us to dive deep with research in machine learning and in partnership with a company that really understands immunology — to try and create something transformative," he explained.
The adaptive immune system, which is the body's system for detecting and fighting disease, is superior to anything that is currently available in healthcare, says Harlan Robins, PhD, head of innovation and cofounder of Adaptiv.
A person's immune system is "aware" when that person is getting sick, and it could react days, months, or even years before a diagnosis is made, he pointed out.
In cancer, the goal of screening is to detect the disease at its earliest stages, when it may be easily treated and is potentially curable, Robson continued. Current cancer screening modalities that aim to detect cancer at these earlier stages have limitations and are only available for cancers at a few sites. For example, with ovarian cancer, there is no screening test, and it is often diagnosed at a late stage. It is one of the diseases that Adaptive is focusing on.
"Ovarian cancer is an unmet need," said Robins, "We know we can treat it if it is detected early, and this is a great area to improve patient outcomes. We are beginning research in a subset of women who are BRCA mutation–positive and who are at high risk of developing this cancer."
Other Tech Companies Moving Into Healthcare
Microsoft is not the only large tech company that has moved into healthcare. The best known is IBM and its supercomputer Watson, as previously reported by Medscape Medical News. Others are also stepping into the healthcare arena, including Google, Amazon, and Apple, along with a growing number of smaller companies.
Recent developments include a Google algorithm that can diagnose diabetic retinopathy in images with accuracy similar to that of board-certified ophthalmologists.
Another example is Apple's partnering with Stanford Medicine to conduct the Apple Heart Study, which uses heart rate sensors in the Apple Watch to collect heartbeat data and to notify users if something is awry.
Amazon has entered the healthcare industry in multiple areas. In 2014, it partnered with Cardinal Health and is now licensed to distribute medical supplies to providers in 43 states. Amazon, JPMorgan Chase, and Berkshire Hathaway have announced the formation of a healthcare company that provides employees with technology solutions to access quality care at a reasonable cost, free of profit incentives.
It is not all smooth sailing, and many problems remain to be to be tackled.
Among these are regulatory hurdles, Lee commented. "How do we get regulatory approval for machine learning? This is all so new."
He commented that, because of the enormous amount of new medical knowledge, AI and machine learning are now absolutely essential. "But at the same time, we're keenly aware that the statistics of cancer are challenging for machine learning," Lee said.
There has to be interplay between current medical research and machine learning, he continued. Medical research is based on the understanding of chemical pathways and cellular mechanisms, and such knowledge has to be coordinated with what can be done with machine learning. "We just can't come in and start processing data," Lee said. "The human race is just too small at this point — at least, that's our point of view."
Lee emphasized that they are trying to approach this with humility. "When you talk about cancer, there are thousands of the smartest people on the planet who have devoted their lives to curing cancer," he said. "We're extremely reluctant to come in and say we have all the answers — we don't. We are trying to understand where we can add value, who are the best people we can work with, and who is willing to work with us."
Health Solutions From Our Sponsors
Source: How Microsoft Aims to Help 'Solve the Problem' of Cancer - Medscape - Nov 28, 2018.