Aivas: a new supercomputer to support AI software deployment in healthcare

The high computing specifications and memory architecture of ALAFIA’s Aivas supercomputer enables intensive AI applications to run smoothly within the single node system. Credit: ALAFIA / Business Wire

The use of AI and generative AI (genAI) applications to speed up diagnosis and unlock new insights for medical imaging is rapidly becoming the norm. Underpinning these applications are the supercomputers needed to run this sophisticated software.

In August 2024, ALAFIA announced the general availability of Aivas, an interactive high-performance personal supercomputer designed for the deployment of AI software applications in healthcare.

The Florida-based startup has stated that Aivas is intended to address market demand by offering the highest compute density and energy efficiency for low-latency, mission-critical workflows and secured workloads in healthcare fields, including digital pathology, neurology, genetics research, and radiology.

The supercomputer has seen early adoption by researchers from the Stanford Medical School Department of Neurology and Neurological Sciences and the Acacia Clinics.

In performing cortical reconstruction using Aivas, Dr Danielle DeSouza, vice president of research at Acacia Clinics and research staff at Stanford University’s neurology department, said the supercomputer helped to reduce the time spent on individual subject reconstructions from over 24 hours to under two hours per subject. The supercomputer was also capable of parallelising to perform hundreds of subject reconstructions at a time.

ALAFIA recently won two innovation awards at this year’s Consumer Electronics Show (CES) technology event in Las Vegas, Nevada, for Aivas.

According to a report by GlobalData, every segment of the AI market in healthcare is set to grow over the next decade and increase from being worth $103bn in 2023 to exceeding $1tn.

Medical Device Network sat down with ALAFIA CEO Cam Buscaron to learn more about Aivas and the healthcare protocols it helps to enable.

Cam Buscaron: Since Aivas is more a B2B product for the likes of hospitals, cancer centres and biotechnologists than a consumer product, the awards recognition from CES was quite unexpected to me.

Due to the recognition, situated in the digital health segment of the booth floor at the conference, we were visited by individuals ranging from a large luxury goods multinational undertaking pathology for cosmetics to doctors and physicians with an interest in health tech-related technology – not the usual CES crowd.

In general, at CES it was interesting to see the degree of enterprise-level foresight companies are displaying towards what their level of AI investment is going to be, which likely wasn’t the case in the past. It now seems clear to me that medium and even small-sized companies are looking into how they can leverage AI, with a concerted budgetary commitment to adopt AI, with these companies of all sizes looking for the right solution they would be to leverage over a number of years.

Cam Buscaron: ​​​​​​​We designed Aivas from the ground up to be the best AI inference appliance, meaning that someone who has a foundation model can run it locally and securely encrypted on our system.

Aivas is a more powerful computer, but it’s also a computer that allows users to do things they wouldn’t be able to do on a normal computer. It opens the capabilities of processes users can perform with large datasets and models. Mostly, this ability comes down to Aivas’s memory architecture. With more memory capacity, bandwidth, and throughput, we can run applications like very large foundation models; where normally you would need two servers in the cloud, such applications can run well in our single node system due to its high-performance specifications.

For a field such as radiology, often it is not just about running foundation models, but about running a number of computer vision algorithms for performing workloads that include organ segmentation and lung nodule detection – all of which perform well on Aivas due to its 206 central processing unit (CPU) cores and large memory provision.

Lung nodule detection is traditionally viewed as particularly challenging, as in a large dataset of over 100 scans, a lung nodule may only appear in two or three. Therefore, a system that can run through these difficult, intensive use cases for radiology, at speed, is highly advantageous.

Cam Buscaron: Most of our customers at this time are in digital pathology. Traditionally, a pathologist would take a tissue sample, create a glass slide and examine it under a microscope. But with digital pathology, the pathologists use whole slide scanners. Placing a glass slide under these scanners, an image is taken of the tissue sample that may be as large as 100,000px × 100,000px.

The benefit of the data gleaned from such large images is that AI models can identify things the pathologist may not be able to see with the human eye alone. In this way, even basic operations, like cell segmentation, cell classification, and nuclei detection can now be done in a faster way to, for example, diagnose the extent of a patient’s cancer in a tissue sample.

In pathology, there are now around two dozen analytics software applications, and we are sort of in the sweet spot between the scanner manufacturers and the makers of this image management software. We partner with the scanner makers for interoperability, and our supercomputer helps to accelerate the processing and analytics capabilities of these applications.

Cam Buscaron: ​​​​​​​We incorporated the company and started working on Aivas around 18 months ago. We were at one of our supplier’s trade shows in March and this was the first time there was some awareness out there about us.

We only properly commenced marketing launch activities when we showcased Aivas at the 2024 International Society of Molecular Biology (ISMB) conference in Montreal, Canada. By this time, we had a website up and running and distributed our first press release about a month after ISMB.

So far, the reception to Aivas has been phenomenal, and the CES awards have helped generate further interest in our product. We’re still very early days. I don’t think who our product is most suitable for is fully aware of it yet, but with growing word of mouth and our recognition at CES, we are beginning to see an uptick of interest in what we have to offer.

As interest does increase, we are thinking closely about our mass manufacturing design and increasing our supply capabilities so that we are well-positioned to ensure customers receive their system in a timely fashion when they order it.

“We do this all virtually on the computer, so we can make the osteotomy in multiple different places to decide where the most appropriate place to do the correction is.”

From here, relevant standard orthopaedic plates are selected for use in the surgery.

Following these preliminaries, surgical guides, jigs, and plastic models of the patient’s anatomy, in this first case the radius, are 3D printed and then sterilised for use in surgery.

“We make sure that the guide fits the bone in the patient exactly the way we planned for it to fit on the plastic bone. Once we have made sure that’s the case, we secure the guide to the bone with wires, and then we do whatever the plan has been,” says Lattanza.

In osteotomy, such plans generally involve drilling holes and then making the necessary bone cuts.

The great thing about this approach, Lattanza states, is that all that needs to be done to ensure the correction has been completed as planned during the surgery is to line up those holes.

She explains: “If the bone is rotated off 90° and when we drill those holes, they’re off 90° on the bone, we make the cut then we rotate and line up those holes to put the plate on because the plate holes are straight, and that’s how we know that we’ve got the correction.”

Beyond making relatively common osteotomies more accurate, a 3D provision also allows for more complex cases to be worked upon. Lattanza relays a recent case in which a child had broken the radius and ulna bones in their forearm.

“During the time that she was growing, this deformity got ‘very 3D’, meaning it was off in the sagittal, coronal, and axial plane,” says Lattanza.

“You can’t see the axial plane on an X-ray, and if you can’t see it, you can’t correct it.”  

In this case, the procedure required two cuts in the radius to restore it to normal anatomy, and one in the ulna.

“In my career prior to having the 3D technology, that’s something that is difficult or impossible to plan and to execute in the operating room, because you wouldn’t even be able to see that you needed two cuts to make it normal again,” explains Lattanza.

Lattanza is keen to add that the influence of 3D printing on preoperative planning and during surgery should not be a cause for complacency, particularly given that there remain limitations to 3D visualisations of CT scans, chiefly in that the current technology cannot show soft tissue.

“Some people think that this is kind of a phone it in now, but that’s not how it works,” she says.

“This is a collaboration between an engineer and a surgeon, and it has to be that way to get a good result.” 

Australia could be one of the main beneficiaries of this dramatic increase in demand, where private companies and local governments alike are eager to expand the country’s nascent rare earths production. In 2021, Australia produced the fourth-most rare earths in the world. It’s total annual production of 19,958 tonnes remains significantly less than the mammoth 152,407 tonnes produced by China, but a dramatic improvement over the 1,995 tonnes produced domestically in 2011.

The dominance of China in the rare earths space has also encouraged other countries, notably the US, to look further afield for rare earth deposits to diversify their supply of the increasingly vital minerals. With the US eager to ringfence rare earth production within its allies as part of the Inflation Reduction Act, including potentially allowing the Department of Defense to invest in Australian rare earths, there could be an unexpected windfall for Australian rare earths producers.