Cloud Computing & Intelligence-Based Medicine
Healthcare Cloud Computing
Dr. Anthony Chang recently authored the book Intelligence-Based Medicine: AI and Human Cognition in Clinical Medicine and Healthcare. As met when he took my cloud computing class at Stanford, he asked me to write a brief on the role of cloud computing in medicine. Below is what I wrote. If you’re interested in learning more about AI in medicine check out the book.
You can’t go thru any airport today or watch any sporting event without hearing about “the cloud”. So what is cloud computing? Quite simply compute & storage cloud computing is the management of the security, availability and performance of compute and storage. Managing security insures the right patch levels are applied. Managing performance insures that a predictable instance of computing is available and by managing availability a cloud computing service insures there are three copies of the data stored. By centralizing and standardizing the management of compute and storage the management can be automated, thereby not only decreasing costs but also increasing quality.
Compute & storage cloud services have enabled new business models. Rather than purchasing hardware up front, locating a place to house it and hiring people to manage it you can buy managed compute for an hour. These economics open up totally new use cases including making high performance computing available to small teams with minimum investments. At the beginning of cloud computing I launched a class at Tsinghua University, one of the most famous universities in China. The Amazon team gave me $3000 worth of cloud computing time. I told the class at the time $3000 would buy you a computer in Northern CA, Ireland or Northern Virginia for 3 ½ years OR 10,000 computers for 30 minutes. It was an example to inspire them to think not about how we do computing today, but what might be economically possible in the future. Of course, today it would be even more computing for even less money. So what might be possible? How could cloud computing reshape healthcare?
The dramatic changes occurring in the application of AI technologies, in particular deep learning, are based on the confluence of three powerful forces. The first is the advent of large amounts of computing, for little cost. See above. The second is the arrival is software for building many kinds of neural networks. Neural network technology has been in the labs for a lot of years, but it is only recently that software such as Tensor Flow or Pytorch has become available. And finally, with the arrival of the Internet we have large amounts of data. Facebook facial recognition can work so well precisely because they have lots and lots of photos of people’s faces.
The confluence of these three powerful forces shown above is propelling a dramatic ability for machines to be more accurate. A simple case study is found in the ImageNet competition. Begun in 2011, the ImageNet contest pitted man against machine to determine who could recognize images more accurately. In the first contest the machines could only be 75% accurate versus humans who were 95% accurate. In just four years machines hit 97% accuracy. Advances are starting in healthcare. Researchers at Stanford University had four radiologists annotate 420 chest X-rays for possible indications of pneumonia. Within a month, the team had developed CheXnet, which outperformed the four radiologists in both sensitivity (identifying positives correctly) and specificity (identifying negatives correctly). While this is a good beginning, it’s just a beginning. There are at least three challenges and opportunities in front of us.
Connecting healthcare machines to the center cloud.
The Stanford university project used 420 chest x-rays. You can easily guess there are 100,000 or more chest X-rays being taken every day. Unfortunately these are locked away in isolated X-ray machines. What if they were all connected and what if we could bring years worth of chest X-rays from around the world? Can you imagine the accuracy of a pneumonia digital assistant? And what if every MRI scanner, gene sequencer, ultrasound, blood analyzer, cell plate reader… was connected? While we have the compute & storage cloud services we need the data to take the next big step.
Analytic Applications.
Center cloud computing is not only bringing low cost compute and storage but it’s also allowing us to build more powerful applications quicker. Up until now most enterprise software companies have focused on building workflow applications and improving back office operations like purchase-to-pay, order-to-cash or hire-to-fire. In the world of analytics we have supplied people tools. Building modern analytic applications require at least 16 classes of software tools. In the old days much of this would have to be functionality you’d have to build. Instead today with the vast amount of open source (e.g. TensorFlow, Hadoop, Kubernetes, Kafka, Databricks, Cassandra, D3, Django, Puppet) as well as variants also available as managed cloud service. It’s now possible to stand on the shoulders of giants and build healthcare analytic applications. Makers and users of healthcare machines (blood analyzers, MRI scanners, ultrasounds, gene sequencers) don’t build their own financial, HR or purchasing software instead they purchase packaged enterprise application cloud services. Perhaps in the future rather than buying tools and hiring data engineers, DevOps people and ML experts you’ll be able to purchase packaged healthcare analytic application cloud services.
Global applications for global health
Once an application is available as a cloud service, whether that’s the ability to purchase a book or providing a pneumonia digital assistant then it has the potential to be available to everyone around the globe instantaneously. By the year 2040 25% of the globe’s population will be in Africa. There is no way to be able to build a first world healthcare system. There is not enough time or resource to build the medical schools, train the people, and construct the traditional hospitals. Instead we have the opportunity to build new applications that can be deployed around the world to analyze, diagnose and prevent diseases. Software technology has near zero costs. Its only requirement is the energy and creativity of people. Many students today don’t want to work on the next dating website or social network, imagine if we could bring together their creativity along with the economics of modern compute & storage cloud services.
Maybe cloud computing can change the world and all those ads in the airport will really have predicted the future.
