Created in 2015, Anatoscope develops software solutions to create 3D avatars from patient medical images. Using this digital twin, dental and orthopedic devices and prostheses can be tested virtually before being fitted to the patient. Anatoscope is a healthtech startup which has grown out of French academic research (CNRS Montpellier, INRIA and CHU de Grenoble).

Frederick Van Meer, COO at Anatoscope, explains: 

“We work in the medical field and, as such, we must host our health data on secure HDS servers. We had looked at an accredited health data host but, given the cost of data storage, it was not feasible. In addition, we wanted our solution to be available via streaming, so the quality of the streaming and the possibility of accessing it from anywhere in the world were essential.”

Issues:

• Making high-performance software streamable
• Ensuring the quality of the service and the streaming in real time
• Guaranteeing data security while controlling hosting costs

"The quality of our streaming is a big issue," comments Frederick Van Meer. “When you stream a 3D view of the design for medical equipment, you can't afford for there to be a lag. Our challenge is to be able to disseminate our software to everyone very quickly, without users having to install it. Our difficulty is that it is used via a web browser and not an external application. The number of embedded technologies is therefore limited.” Challenge overcome: "Thanks to our infrastructure, we can stream high-performance software, which does not require a powerful local machine; a simple PC or a tablet is sufficient".

Anatoscope has created its own streaming architecture, based on Amazon S3, EC2 GPU instances, Docker, and a load balancer. “We use a lot of GPU instances because our applications are computer-aided 3D design software. They are deployed on our clusters according to user demand, whether we have one or ten, knowing that one case can use a full GPU instance. The GPU is 70 to 80% of what we do, and the responsiveness of AWS to peak computing demands was an important factor in our choice.”

Another challenge to overcome: guaranteeing the security of health data. The hosting of health data must meet specific security requirements. “With streaming, firstly, the data always remains on our server, i.e. on AWS. Secondly, when they were accredited as an HDS, we repatriated our data which was with another HDS host. The storage costs were unthinkable for us. We started with a storage cost of 55 USD/GB; by migrating to Amazon S3, the cost reduced to 0.023 USD!”, notes the COO.

This migration allowed them to simplify their organization, “What a mess! We used to store the patient identification data with an HDS host and then encrypted it to send to AWS for a limited time."
In five years, Anatoscope has evolved in terms of volume, size and activity. While the architecture as such has not been transformed, its deployment and the tools required have changed significantly since recruiting a DevOps skilled in using AWS. “We did a lot of things by hand and we always had to be there to prevent any problems. Now we mostly save time - for a given application - when duplicating environments (like creating beta or prod from work done in alpha). We went from a few days to a few hours!”
The automation relies on tools external to AWS, as well as AWS architecture elements: VPCs isolated by customers/partners, automated AMI builds and images deployed via a LaunchTemplate for their GPU instances, the management of these GPU instances via an AutoScaling Group, with Lifecycle and Schedule managed by AWS Lambda and SNS. On the API/Web instances, automatic deployments also manage DNS records on the anatoscope.fr domain via Route 53, and web certificates via ACM.

“The great advantage,” notes Frederick Van Meer, “is also being able to duplicate our AWS infrastructure very easily (mainly in Europe/Ireland, currently), and with a practically equivalent cost, in other regions of the world to meet our partners’ needs. For example, a project is underway in the United States.”

As a member of the AWS Activate program, Anatoscope has received credits, which they have used to test their infrastructure. “Without them, we might not have tested some things. We would have been more cautious about spending thousands of euros on a solution that might not have been the right one. Being supported by AWS has helped us a lot with technical choices.” Not to mention the "great ease of quickly testing architectural concepts, and the performance of tools for the integration of our in-house solutions", as other criteria which promote testing.

In fact, lockdown highlighted the resilience of the software solutions offered by Anatoscope. "Users were able to train in the software remotely, without difficulty, by simply logging into the platform, as if they were at their desks, because our software is online and we have no problems scaling according to the number of users,” comments the COO.

This infrastructure also facilitated the rapid deployment of new applications. “In the context of COVID-19, we implemented AnatoMask, a solution that allows you to customize a mask in 3D on a user's face. It scans your face, sends your data, and we generate a blueprint for a custom 3D printable mask,” explains Frederick Van Meer. The startup adapted one of its applications for creating helmets for children with plagiocephaly. “We developed it and made it available within a few days, which would have been difficult without the AWS servers because the software would have had to be installed at the users' premises. This way, it is sufficient to create an account for them, and a clinic can, for example, autonomously create 3D masks for its employees.”

“Thanks to AWS, we have succeeded in developing an infrastructure on three types of software. In dentistry, the software is used daily by various prosthetists. In orthopedics, we have succeeded in signing with a world leader. The next step will be to address the scalability of our solution with the increase in the number of users. Big data management is also part of our future challenges: how to store a large volume of medical data, at a lower cost, over the next five or ten years. Our clinical cases are between 300 and 400 MB in size, and we will have to think about a long-term storage solution (Amazon Glacier perhaps).”

"With regard to our technical choices (making high-performance software available via streaming), we realized with COVID-19 that we could come up with a solution that was very easily accessible. As we were focused on our orthopedic and dental applications, we didn't even realize. We have proven once again that our software can be used without difficulty anywhere in the world, at home, in the office, or elsewhere.”

• 3 days to develop the 3D mask customization application
  
• Reduction in costs from $55 to $0.023/GB of storage by migrating to Amazon S3
  
• Reduction in time from a few days to a few hours to duplicate environments