Healthcare AI in the Cloud

It would seem to be the understatement of the century to say that AI is a big deal in healthcare. Even though controversy swirls around the data science tools and techniques collectively known as machine learning, their impact on the global conversation about improving healthcare is undeniable. That’s especially true for those who have been involved in the intersection of health and technology in the last few years. For reasons ranging from changing population and economic dynamics, to the widening inequality in health outcomes, to the emergence of pandemics, there is little argument that we need to get more insights from the oceans of available health data. Those insights can be used to improve outcomes, increase the quality of life for clinicians and patients alike, and lower healthcare costs. These goals are known as the Quadruple Aim.

Fortunately, machine learning is a terrific way to analyze vast amounts of existing data and to extract insights from new data that the AI models have not been exposed to before. Regardless of the technical details, such as the types of artificial neural networks we use in our models and how we train them specifically, machine learning is the perfect choice for AI applications in healthcare.

What do we need in order to achieve Quadruple Aim goals with AI? First and foremost, data. Lots and lots of data. Then, a lot of computing power to train our models with that data so that we get the highest accuracy possible with the lowest error rate. Finding the right balance between levels of sensitivity (true positives) and specificity (true negatives) is important as well. This process adds training runs, further increasing the need for compute and storage.

Where do you go when you need lots of storage and compute resources that you can easily spin up and down whenever you need them? "The cloud," of course. Public cloud providers were built for this purpose, offering virtually unlimited compute and storage resources as a service, with a wide complement of tools that you pay for as you use them.

This speed, elasticity, and scale are why many healthcare AI projects start in the cloud. What happens next is interesting. An AI-centric organization’s needs can grow from terabytes to petabytes (or exabytes) of storage for training data, and from a few GPUs to hundreds or thousands. Conventional wisdom says that leasing these resources from cloud providers becomes more expensive than buying and running them on premises.

This view is based on outdated assumptions that don't take NetApp's services for cloud compute and storage into account. For the last 10 years, NetApp has been on a journey to transform itself into what it is today: a cloud-led, data-centric software company. This means that we help store, manage, protect, and process data in the best, most cost-effective way no matter where that data is located—at the edge, in the core, or in the cloud. This location agnosticism allows us to do what is best for our customers without being saddled by legacy thinking about where data should reside. Offering NetApp® ONTAP® data management software as a service in the public cloud changed everything, upending conventional wisdom.

Healthcare AI projects often use unstructured file data, such as medical images, free-text visit notes, procedure reports, and recorded conversations. NetApp's native file services, such as FSx for NetApp ONTAP on AWS and Azure NetApp Files, offer a fully managed experience in the public cloud. Added benefits include ONTAP data efficiencies, performance enhancements such as FlexCache® software, and security features such as SnapLock® compliance software. These benefits  reduce the overall cost of storage in the cloud while delivering the scale and performance that AI projects require.

Additionally, we help data scientists reduce the time they spend wrangling data by making API integrations available between ONTAP and data science development environments such as Python, Jupyter Notebooks, and Kubeflow. You don’t need to make compromises or refactor your pipelines to move seamlessly between the edge, the core, and the cloud. With ONTAP as a service, you can have it all.

When training AI models at scale, compute needs increase significantly. Whether you’re using CPU or GPU instances in the cloud, you can keep the associated cost under control with Spot by NetApp. Spot is a suite of cloud services that use AI to drastically reduce the price of compute resources in the cloud. It can automatically optimize your cloud portfolio, helping you fully use commitments and manage commitment lifecycles to align finance and DevOps. Spot can help you control costs by using spot instances with guaranteed availability for savings of up to 90%.

Although there are reasons for preferring an on-premises AI infrastructure, inaccurate conventional wisdom should not dictate your choices. NetApp has been making investments in cloud and data science for many years, and we’re ready to be your strategic partner for healthcare AI projects in this era of hybrid multicloud.

Many data scientists work at NetApp. They understand the problems that other data scientists face and have developed tools to help make their lives easier. Along with our cloud architects and healthcare experts, NetApp data scientists stand ready to augment your teams so that you can use AI along with patient data to achieve Quadruple Aim goals.

If you’re interested in taking the next step, contact me to set up a conversation. One of our Cloud Design Workshops could be a good next step.