You are reading 1 of 2 free-access articles allowed for 30 days
Priscilla Lynch speaks to Dr Pearse Keane about his pioneering work in the field of artificial intelligence in medicine
Artificial intelligence (AI) will very soon revolutionise the way patients are screened, diagnosed and treated. The specialty of ophthalmology is leading the way in unlocking its potential in the medical arena, according to one of the pioneers in the field.
Irish doctor Dr Pearse Keane, Consultant Ophthalmologist at one of Europe’s leading ophthalmology centres, Moorfields Eye Hospital, UK, has led the development of a ground-breaking AI algorithm that can identify indicators of over 50 sight-threatening retinal diseases and which recommended the correct referral decision with 94 per cent accuracy, matching or exceeding world-leading ophthalmologists, in a proof-of-concept trial.
The ground-breaking AI algorithm came about after Dr Keane, who studied medicine in University College Dublin, reached out to Google’s DeepMind team in 2015 to investigate whether AI technology could help improve the care of patients with diseases such as age-related macular degeneration (AMD) and diabetic retinopathy.
A formal partnership was launched in 2016 and brought together leading ophthalmic professionals and scientists from the UK’s National Institute for Health Research (NIHR) with some of the UK’s top technologists at DeepMind. Over the course of the five-year research project, Moorfields, which carries out approximately 5,000 three-dimensional macular optical coherence tomography (OCT) scans a week, will share approximately one million de-personalised OCT scans with the researchers.
To create the AI algorithm the research team applied novel deep learning technology to clinically heterogeneous OCT scans. Using two types of neural network the AI system quickly learnt to identify 10 features of eye disease from highly complex OCT scans. The system was then able to recommend a referral decision based on the most urgent conditions detected.
To establish whether the AI system was making correct referrals, they benchmarked the results, so that clinicians also viewed the same OCT scans and made their own referral decisions. Their study concluded that AI was able to make the right referral recommendation more than 94 per cent of the time, matching the performance of expert clinicians.
This AI system can also provide information that explains how it arrives at its decisions, which is critically important for clinical application, since it will allow doctors, nurses, and healthcare professionals to scrutinise these recommendations, Dr Keane told the Medical Independent (MI).
The research also demonstrated that this particular AI technology can be easily applied to different types of eye scanner, massively increasing the number of people across the world that it could benefit, and future-proofing the technology against new devices and models that could emerge in the future.
Having published the proof-of-concept data on the AI algorithm in Nature Medicine in 2018, Dr Keane is keen to see the findings implemented in real life and he hopes that a clinical model will be ready to roll-out within the next couple of years.
“It’s like as if we built a concept car that can break the world land speed record with a team of scientists and engineers ready to tweak it, but now what we’re thinking is ‘how do we make a production car?’ How do we rebuild the algorithm from the ground up so it can do the same thing, but with a fraction of the computing power in a fraction of the time and can be used by ophthalmologists all around the world? We are well on the way to getting there and hope that will be there in the next couple of years. I don’t think that it is five years away or longer; this is something that is coming soon.”
He pointed to the 2018 US Food and Drug Administration approval, under a breakthrough device classification for the first diagnostic system to use AI to detect diabetic retinopathy in adults who have diabetes, as paving the way for a new wave of AI-based technologies in medicine.
While AI’s usefulness for screening for particular eye diseases has now been proven, this is only the beginning – the potential applications of AI in medicine are wide-ranging and will one day assist in screening and triaging as well as diagnosing, treating, and predicting outcomes in patients in everyday clinical practice, he said.
Potential uses of AI include helping predict the future progression of diseases like AMD in order to help with patient risk stratification or assessing treatment efficacy, Dr Keane told MI.
“So seeing if patients are getting better or worse, for example, if they are receiving intravitreal injections for AMD, or diabetic macular oedema, or drops for glaucoma, to assess whether they are working.
“Despite all the hype around AI, it does really have the potential to transform healthcare. Ophthalmology is going to be the first of all the medical specialties to be fundamentally transformed using AI and I am very excited about that,” he said.
For this to happen, however, clinicians must be at the forefront of developing and validating these new technologies, Dr Keane maintained, referencing his recent article in The Lancet Digital Health on the topic.
“Our systematic review concluded that, looking at the quality of most AI studies at the moment, they are mainly written by the machine learning world and the quality of the clinical validation is not that strong,” he said.
“So we argued for proper clinical validation and proper clinical trials of these algorithms.
“I firmly believe for AI to be successful in medicine and a specialty like ophthalmology, it has to be driven by healthcare professionals. I think if you empower ophthalmologists we’ll come up with hundreds of applications for AI systems and be able to assess them.”
Furthermore, ophthalmology could be “an exemplar” for other medical specialties in the development, validation, implementation, and adoption of AI, according to Dr Keane.
“So in other words, an exemplar in terms of identifying cases where AI can add value to patients, clinicians, and healthcare institutions; and in setting up infrastructure where AI can aggregate and curate data for the development of AI systems; as well as the validation of these systems through randomised clinical trials that will properly demonstrate these things work and are safe and effective and bring benefits to patients,” he said.
Replacing the doctor?
However, Dr Keane believes it will be some time “before someone makes the diagnosis in the patient and gives them an injection or surgery or laser fully by AI systems, as I don’t think they are ready for that”.
He also acknowledged that rolling out AI in medicine will have to be done in an incremental, evidenced-based manner.
“I think we need to balance enthusiasm and excitement for these new technologies with caution. They do have the potential to be transformative, but there are lots of ways they may not work or go astray. We can still be enthusiastic, but at the same time demand very high standards before we use them on patients.”
One of the key potential barriers is suspicion that AI could eventually replace clinicians in some areas or operate independently without oversight.
“Nobody is going to be replaced, which is a concern for some when they hear about AI and no one is suggesting that we just let these systems loose on patients to do very important things such as decide whether they need an injection or surgery without a doctor/healthcare professional being in the loop. So the technology is still very young and we are trying to find the areas which can bring the best benefits for patients,” Dr Keane stressed.
He emphasised that AI will be just another, albeit very useful, tool for clinicians, against a background of ever-increasing demand for healthcare services.
“Does anyone really think that in five or 10 years’ time our clinics are going to be anything other than super busy? When we solve one problem, another pops up and given the constantly increasing patient demand, as patients live longer and so on, everybody recognises that this is not a luxury to develop these technologies; it is a necessity.”
Dr Keane will address the 2020 Spark Summit in February on the topic of AI in medicine, where he will discuss his work in the field to date, including his latest study, published in The Lancet Digital Health, which used an automated deep learning platform that allows people without any coding experience to train AI models themselves.
“What we were able to do is get two members of my research group who don’t have any coding experience at all and they were able to get medical imaging datasets; chest x-rays, skin cancer scans, OCTs, etc, and within a few days they were able to get results that were comparable to state-of-the art AI,” Dr Keane said.
“To me that is a very important thing as it signals that we are moving towards an era where there is to be democratisation and industrialisation of these AI systems. So it will be less about people with very specialised expertise in AI with huge computing resources, but instead about people with domain expertise and good data who are developing these algorithms. So the analogy that I would use is that it will be like the late 1970s when Apple created the first home computers and previously they had just been used by big corporations for things like payroll… but they realised that when you give people these personal computers and the ability to explore them, they will come up with hundreds of applications. So I think these automated AI platforms could have as transformative an effect.”
The second HSE Spark Summit takes place in the Convention
Centre, Dublin, on 13-14 February and will focus on the application
of AI and robotics in healthcare, the benefits of big data and personalised
medicine, emerging medical devices, digital disruption, and the future of
health education. Full details at
Dr Vincent Maher, Consultant Cardiologist, Tallaght Hospital, Dublin, told delegates that there could be up to...
An update to the Covid-19 vaccination programme as a result of the threat posed by the...
There is “no central collation” of data on Covid-19 staff derogations during the pandemic in either...