Main content

Medical start-up uses AWS machine learning to turn 2D patient scans into 3D printed models

Gary Flood Profile picture for user gflood May 20, 2022
axial3D is using AWS to build 3D models that are helping doctors reduce time in surgery and improve the patient experience

Image of axial3D models of spine and chest
(Image sourced via axial3D )

A Northern Ireland healthcare SME is using a wide swathe of Amazon Web Services (AWS) technologies to enable its proprietary algorithms to convert CT and MRI scans into three dimensional models.

The firm - axial3D, based in Belfast - is making use of AWS’s SageMaker Machine Learning model software, Lambda serverless computing tool, Step Functions microservices orchestration tool, an RDS implementation of the MySQL relational database, and S3 File Storage.

Speaking at the recent AWS Summit in London, the company said the combination of its original image research and AWS Cloud means it can better support clinicians across the entire patient's care pathway with much better imagery than ever before.

Roger Johnston, the Medical 3D firm’s CEO, said the solution addresses an issue that often hides in plain sight: that even medical experts often find the flat black and white imagery, which we are all used to seeing in the Doctor’s surgery, as too opaque. Johnston said: 

Most people’s experience of medical imaging is on the receiving end of orthopedic trauma or having children. But when the surgeon says, ‘as you can clearly see’, they actually can't see any better than you. Most often, it’s a narrowing down to scenario B, C or D, and when they open up, they find it’s ‘C’ after all, and go with that.

Johnston says that historically hasn’t been that much of a problem but does mean that medical professionals need to plan for multiple scenarios that reduce their overall time efficiency. However, AI has now got to the point that it can look at a scan and based on all the accumulated learning from several million pre-labeled predecessor images, build up a picture of a problem area or organ - and to a level that all that educated guesswork isn’t needed any more. 

The firm’s claimed pixel identification accuracy is over 99 percent, depending on the algorithm and what specific anatomy is - though the firm stresses that human experts always perform a final check.

Helping spot issues earlier in 3D 

In a recent case, as an example, an infant at Southampton Children’s Hospital had successful life-saving surgery after her surgical team first practised on a fully accurate 3D model of her heart. Turning 2D into 3D also surfaced a previously unseen condition - the patient had a rare anomaly of the heart’s collecting chambers, as well as a duplicate of the main vein inside the upper body. It was later also discovered she had an even further heart defect, total anomalous pulmonary vein drainage.

So useful was the ability to see an accurate representation of these actual organs in 3D that one of the surgeons called it an “incredible piece of technology which can change the way we approach congenital heart disease treatment in children.” Surgeons but also patients and their families benefit from this use of AI and the cloud, he said:

There’s a democratization of medical education here too. Instead of not understanding what the hell I’m supposed to be seeing on the screen, I can hold a model of my spine and show the specific slipped disc that’s causing all the problems in my hand.

Overall, the company claims use of its tool results in 50% of surgeons changing their plan for surgery, with 62 minutes surgery time saved on average, and 18% less time in hospital for every patient.

A key enabler of the evolution of giving medical teams this level of pre-op understanding, said Johnston, has been the healthcare industry’s adoption of the DICOM file format, which is now the default for all kinds of machine-generated imagery, from MRI to PET.

That standardization means that a system like axial’s doesn’t need to worry about variations, so it can be relied on to know what every pixel on an image, from blood vessel to tumor, really is.

Hence the main axial deliverable - that it can turn any 2D medical image given to it in stacks into a 3D image automatically on AWS in minutes. Cloud plays its part by making it very easy for healthcare organizations to access such images, as they don’t need to put any third-party software on to their network but can simply open a representation over the Web.

Johnston, a former management consultant and business development specialist for several software and services companies, says that another advantage of working with Amazon beyond all these technical factors is the company’s approach to SMEs. Johnston said: 

There are all types of technical reasons to use them, they've got more MIPs and more gigabytes than anyone else, from access control to EPA compliance, and all the modules are just utilities for us. 

In terms of features and functions, the utilities available on the platform are more comprehensive than anyone else; Microsoft would be a relatively close second. But in terms of data, privacy, data, compliance, medical regulations - Amazon, again, has a more complete set than anyone else.

But more importantly, instead of them giving my CTO every tool under the Sun and going, ‘Good luck,’ what they'll do is tell you what the right architecture is. Amazon will encourage you to only buy the bits you need, as opposed to trial, and error and reinventing.

Finally, Johnston and his team also claim that even though axial is still “a tiny little ISV,” Amazon is also very supportive in terms of co-marketing and commercial partnership. He added:

Other cloud firms give you a start-up no end of technology accessibility credits, but they also have hundreds of partners and start-ups can end up at the bottom of the priority list. But Amazon does just one thing, cloud, that's what they do, and therefore they are much more focused. 

We were Amazon’s global healthcare partner of the year two years ago, for example, because they know the potential we have in this market. We’re a five-person company, not a 50-person one - but if I want to meet GE, Philips and Siemens, Amazon will give me access to all of those, and as I’m going in with them, I’m going in with credibility. 

We have compressed the market reach stage to months, not the years it would have taken with other cloud companies.”

In terms of next steps for the brand, Johnston said that its solution is already being used in 50 NHS Trusts and 250 hospitals worldwide, with an aim of making the service as cheap and globally available as possible.

A grey colored placeholder image