AI Software Writing AI Software For Healthcare?

At the World Medical Innovation Forum this week, participants were polled with a loaded question: “Do you think healthcare will become better or worse from the use of AI?”

Across the respondents, 98 percent said it would be either “Better” or “Much Better” and not a single one thought it would become “Much Worse.” This is an interesting statistic, and the results were not entirely surprising, especially given that artificial intelligence was the theme for the meeting.

This continual stream of adoption of new technologies in both clinical and post clinical settings is remarkable. Today, healthcare is a technology operation. As a case in point, outside of the array of MDs and medical professionals presenting at the forum, there was clearly a strong, advanced technology thread weaved throughout the conversations of the traditional topics of pathology, radiology, bioinformatics, electronic medical records (EMR), and standard healthcare provider issues. As an example, a panel of senior technology experts from Microsoft, Cisco Systems, Dell EMC, Qualcomm, and Google joined research and information officers from Partners Healthcare and Massachusetts General Hospital to discuss the challenges in what they called “Data Engineering in Healthcare: Liberating Value.” That is a serious title for a panel.

Data portability was clearly a key topic, as was security and the public cloud. The underlying issue with the cloud is that the EMR was never really designed to be portable. Health records existed with institutional walls, and were not originally intended for real time care, but more as a means of tracking costs and transactions as the patient traveled through the various systems. As the EMR has not only become more feature rich, the ability to mine that data inside of them with ML and AI methods is clearly at the forefront of everyone’s mind right now.

There was discussion of episodic systems wrapped in policy and technology – this really isn’t quite how we can gain the maximum knowledge from the healthcare version of a Digital Me. A digital object containing all of our many and varied health related attributes. The challenges of discussing how to best build a “marketplace” and healthcare data exchanges and how to integrate “data marts” with existing EMR systems was obvious. The value of an EMR record liability was stated to be about $400 per account, and with huge systems such as Partners Health holding over 6,700,000 records alone, the size of the financial challenge and burden is clear, let alone the challenge of performing bioinformatics at the bedside.

A particularly lively discussion also ensued when privacy and the European GDPR was raised, especially the ability to “be forgotten” in the context of the electronic medical health record. Disagreements in approach were clearly observed between the European and US contingents, especially their divergent concepts of their data privacy strategies. The subject of data ownership and blockchain was also front and center of a number of conversations, “who will be Switzerland” was asked. Should there even be a data Switzerland? The edge as we have discussed here at The Next Platform also plays a significant component in modern healthcare delivery. Two networking companies, Cisco and Qualcomm, both explained their importance in the overall ecosystem of ever increasing wireless density, and mobile and security layers for diagnostic and post clinical health care. There are an ever increasing number of “apps for that,” and advanced monitoring appearing in the market, many of which have life sensitive, critical data as part of their output.

Modern Healthcare Is A Tech Operation

In addition to the panel on technology, Jensen Huang, co-founder and chief executive officer at Nvidia, joined Keith Dreyer, the chief data science officer at Partners Healthcare and a professor of radiology. Not that many years ago, this coupling of experts would have been unheard of, but now advanced engineering and healthcare professionals continue to be an all-important combination for human health and outcomes. Huang said at the meeting: “AI is the greatest technology force of our time, it has the ability to achieve superhuman results.” Bold statements indeed. Huang was also keen to point out, as far as he is concerned, “AI is software coupled with your data strategy which then writes new software by itself.”

Software that writes itself? We needed to think about that comment some more, how does this help healthcare? Is the software that the software writes going to be the right software? So many questions.

Nvidia spends more than $1 billion in research and development each year with over 3,000 engineers working on AI in its autonomous driving division, and the company is investing heavily into “software that writes software.” The overarching aim for Huang is to “augment humans,” to provide AI and integrated computational systems to enable the species to make better decisions based on ever more complex data, to effectively turn them into “super human.” To be clear, the entire field of medical AI is an inordinately high stakes game, vast budgets with the potential for massive disruption. The level of industry hyperbole reflects how high the stakes are.

To understand more The Next Platform took some time to sit down with Kimberly Powell, vice president of healthcare at Nvidia. Powell has been on the road show with “Project Clara” since the announcement at GTC explaining to many what the work is all about. Here at The Next Platform, we were unclear, was it about remote imaging, was it about retrofitting older radiology systems, was it some new software, was it about computing in the clinic? So many questions, Project Clara appeared to be all things to everyone, a kind of miracle medical AI, so we wanted to understand what it really was all about.

Virtualization Of Medical Devices

Think back to VMware ESXi, Citrix Systems Xen, and Red Hat KVM hypervisors. When these server virtualization layers were first released into the wild, many thought of them as parlor tricks. Cute, we can now have computers run computers, essentially an early version of software running software. Had we as a community simply stopped there, we would never have seen the advances in public cloud, DevOps, and systems management we need to supply serverless compute and more complex services to the world. We would have missed the point.

Here at The Next Platform, we see Project Clara to be in an almost identical vein to the early hypervisor conversation. It has the potential to virtualize over three million worldwide medical devices where real tangible benefits will be gained. Powell explained that there are three key steps needed in any medical imaging device: Reconstruction, Classification, and Visualization. The strong similarities to the field of autonomous driving were obviously not lost on us. The ability to build on years of development of AI compute coupled with strong and mature software libraries and the ability to slice and dice compute into discrete elements are the key component parts here.

Software Defined Medicine

Effectively the instrument is abstracted from the physical hardware and into software. “Software defined medicine” would be an appropriate classification for these new breed of medical devices. When asked Powell how Nvidia would not end up directly competing with existing medical device providers, they were clear to point out that Nvidia is partnering with the CEOs of the traditional medical heavyweights, and reiterated the quote from Huang by saying that the domain experts are effectively “teaching” the devices to be experts, with the devices then having sophisticated software to insulate the teacher from the underlying computational complexity. DIY kits of parts isn’t the end game either here, much like autonomous driving a set of “certified pieces” will need to be combined to deliver more insight from 3D volumetrics and CT and MRI scanners. This defined the partnership with other providers in the complex medical device ecosystem.

The plan also includes that “certified pieces” may also be combined to construct new hybrid imaging technologies, for example, to look at soft tissue as well as bone and hard features, augmented with insights from data. Powell was also keen to point out that while this looks like a “cloud offering”, the federal requirements for on premise and data security issues are forcing that the same software using the Nvidia GPU Cloud repository to also be installed inside the walls of the institute. This quickly alleviates concerns about data privacy and security.

Once we looked closer at Project Clara, it is essentially all the tricks we have learned since being able to virtualize compute are now also being applied to advanced systems to effectively build new “virtual medical devices”.  It is a brave new world of AI and virtual compute that Nvidia says will deliver superhuman results.

Sign up to our Newsletter

Featuring highlights, analysis, and stories from the week directly from us to your inbox with nothing in between.
Subscribe now

Be the first to comment

Leave a Reply

Your email address will not be published.


This site uses Akismet to reduce spam. Learn how your comment data is processed.