Dr Aaron Golden, NUI Galway. Image: Ray Butler
Dr Aaron Golden of NUI Galway is looking to develop an AI imaging system that can help radiology teams detect the early stages of Covid-19.
When it comes to the treatment of Covid-19 – a disease we are only beginning to understand – time is of the essence. However, getting a diagnosis quickly and accurately has proven a challenge in many parts of the world.
While nasal and throat swabs have become the most familiar tests, they can often face significant delays. Instead, CT scans can be a useful tool because a radiologist will be able to see evidence of lesions on the lungs that would be indicative of Covid-19.
This comes with its own issues, however, as it can be difficult to tell whether a lesion is caused by community-acquired pneumonias, other pulmonary disorders or a host of other common lung conditions. According to NUI Galway’s Dr Aaron Golden, this is where AI could play a critical role in supporting radiologists.
Golden, along with a team of clinicians, is looking to build an AI imaging system that could rapidly detect lesions associated with Covid-19 in minutes. By using a type of machine learning algorithm called a ‘generative adversarial network’, the team aims to correct for variations in imaging protocols between differing radiology facilities.
Siliconrepublic.com caught up with Golden to find out more about the project, ahead of his online presentation at AtlanTec 2020.
‘AI is absolutely critical in processing vast amounts of data and discerning those subtle melodies against a wall of noise’
– DR AARON GOLDEN
How did this project come about?
I co-supervise a PhD student with my colleague Dr Christoph Kleefeld who has a joint position in NUI Galway and University Hospital Galway (also a co-investigator on the Covid-19 project) and is working on using AI to [determine a patient’s stage of cancer] using CT scans.
I had been tracking what medical teams in China had been doing with CT scans to triage Covid-19 patients and I could quickly see an opportunity. I talked it over with Christoph and he could also see the potential.
He reached out to his colleague Dr Declan Sheppard, a consultant radiologist who had pioneered the application of AI techniques in radiology, so we were all on the same hymn sheet. One of the most valuable things about working in Galway is it really is a small world, and things can happen really quickly here once the connections are made.
How would your system differ to other similar AI technologies being used to tackle Covid-19?
The big difference is the way in which we plan to go about standardising the training data involved. Whilst we can access thousands of anonymised CT scans, the key thing is no two CT scans are perfectly baselined.
There are always instrument settings and systematics associated with the scans, things that highly trained radiologists would learn to discount. But these are precisely the things that an AI system would zero in on, instead of the subtle lesions associated with the wide variety of pulmonary conditions that even healthy people have. I guess you could say that’s the secret sauce!
Do you think AI will have an overall positive impact on addressing Covid-19 now and in the future?
I think it absolutely will on many levels, from vaccine design to identifying potential drug targets on the virus’s surface. It will also help in screening thousands of existing approved pharmaceuticals to determine if any possess the ‘right shape’ to bind to these targets.
AI is absolutely critical in processing vast amounts of data and discerning those subtle melodies against a wall of noise, which represent these ‘hits’ we can work with. I’m really interested in fusing image data – such as that from CT scans – with patient genome data to try and join the loops that explain why some patients get really sick, and others shrug it off.
Can we use AI to build up prognostic maps of the patterns of lesions within a patient’s lungs and use this to determine likely outcome, particularly if they recover from Covid-19?
There will be hundreds of thousands of survivors living with complications after a vaccine is found. AI can play an important role in ensuring they get the right treatment options.
With this kind of technology, how do you ensure a patient’s data privacy is maintained?
The endpoint of this project is a trained AI system residing on a standalone, high-end PC desktop located in the clinic. Its only connection will be to the existing radiology imaging suite so it will be ‘within the fence’, so to speak, and part of the secure networks the hospital systems already have in Ireland.
Its sole job is to process a given CT scan against its ‘memory’ and to plot up the results. As regards training, all scan data will be anonymised either prior to collation or else via the use of approved anonymising software.
We are being incredibly rigid in the development of this project. It will be trained on freely available anonymised data, but when used on patient data, that will happen in a clinical environment following HSE guidelines and protocols.
Are there any other potential uses of this technology?
We’re committed to delivering a trained AI system for the Health Research Board. This will be trialled in the radiology department of University Hospital Galway and will, in effect, mark the next stage on the project’s journey.
This is where it starts to be used in clinical practise with the hope of expediting the diagnosis and staging of Covid-19.
However, this is potentially just the tip of the iceberg as the same system could be used to support CT scan radiology of cancers. The imaging modality could be changed so that the same infrastructure could be used with MRI scan data instead of CT scans, widening its clinical application.
Of particular value would be the system’s ability to identify specific structures and patterns associated with disease types and stage. Being able to combine this information with a patient’s genomic data could be invaluable in tailoring precision medical therapies. This project is very much a starting point!
AtlanTec is a multiple-event festival that will take place from 18 to 22 May and will be run on the Cisco Webex conferencing platform.
