Interviews

"Artificial intelligence is an impending force that will sooner or later become an aid for doctors"

 Photo: Jordi Casas-Roma

Photo: Jordi Casas-Roma

17/06/2020
Slvia Oller
Jordi Casas-Roma and Ferran Prados, research leaders of the Applied Data Science group

 

Researchers Ferran Prados and Jordi Casas-Roma are the leaders of the Universitat Oberta de Catalunya's (UOC) research group Applied Data Science (ADaS) Lab. ADaS, which is attached to the University'sFaculty of Computer Science, Multimedia and Telecommunications, devises artificial intelligence-based solutions to help improve people's health and quality of life. Jordi Casas-Roma and Ferran Prados are both engineers with PhDs in Computer Science. Casas-Roma also has a master's degree in Advanced Artificial Intelligence, for which he focused his final project on data privacy and data mining. Likewise, he is presently director of the Master's Degree in Data Science at the UOC. Meanwhile, Ferran Prados specializes in medical imaging and, in 2012, he began working for the Centre for Medical Image Computing at University College London, which he currently combines with his work at the UOC. Among his achievements, Prados has developed advanced methods in image analysis of the spinal cord and the brain. The sum total of their knowledge enables them to conduct research into aspects of medical imaging, genetics, medical data privacy, energy poverty and the evolution of patients with multiple sclerosis based on artificial intelligence tools. They are also currently working on a project, pending funding, to help in screening COVID-19 patients.

 

Photo: Ferran Prados

Since it's at the front of everyone's minds, what does the coronavirus project you're working on consist of?

FERRAN PRADOS (FP): The project seeks to create a single access point for diagnosing COVID-19 that helps in patient screening. In other words, we would like to make the diagnosis on the basis of a website available to all hospitals where any image technician can just upload the resonance, the CAT scan, the ultrasound or the X-ray they've taken of patients with COVID-19 symptoms, and based on artificial intelligence, the doctor is able to find out the probability of the patient having the disease in just a few seconds. It's a tool that aims to help guide doctors in their diagnoses.

Would it avoid the need for testing, for example?

 

FP: It would be in addition to testing. Just as we know that tests can fail, so can algorithms. What the doctor has to do is analyse them together. The doctor might find that the serological test reads negative, while the computer program says that the patient has a high likelihood of having the disease. Based on what the algorithm says, plus the tests or examinations that they've done on the patient, the doctor will decide what the most accurate decision for the patient is.

What benefits does it have for the patient?

 

FP: The benefit to the patient is that they'll get another test and their diagnosis will be more meticulous and specific. But it also has benefits for hospitals as it helps them optimize their resources. The tool helps with triaging patients. For example, by not sending anyone to ICU if it's not necessary or not sending anyone home if that's not the right thing to do either. The tool should help prevent what happened at the start of the pandemic: everyone turned up at the hospitals and no one was really clear on where to send them.

At what stage is the project right now?

JORDI CASAS-ROMA (JCR): We're in the phase of seeking funding because we need machinery resources and people to help us, but in the meantime we're moving forward on the project because the more things we've got in place, the greater the credibility we'll have and the greater the probability of success when we submit to a call for funding. At the moment, we're advancing as far as we can with the resources we've got.

Linked to medicine, you've been working on a project for some time now with IDIBAPS, the research institute at the Hospital Clnic, which will be of great help for patients with multiple sclerosis. What discoveries are you making?

JCR: The study has been running for over a year and is in two parts. On the one hand, we want to see how we can categorize the severity of the disease depending on the integrity of certain regions of the brain. We've determined that there are six or seven key regions in the brain and we've seen that when the disease advances more, these regions have suffered significant damage and have ceased to be information transmission centres in the brain. The second part of the study consists of predicting a patient's state or evolution, in other words, the path that someone with multiple sclerosis will follow. In the future, this may lead to greater personalization of pharmacological treatments according to the evolution of the disease.

FP: On the basis of the regions of the brain and how they are interconnected, we can create a connectivity map. Depending on the patient's stage of the disease, these connections steadily degrade. We can detect the changes that take place in the connections of the 76 most external regions of the brain and classify patients according to the cognitive loss that they experience. But we don't to leave it at that. Besides knowing how the different areas in the brain connect, we also want to know how the patient responds to external stimuli: for example, we want to know what areas in the brain are activated or deactivated when we ask them to move a finger or look at an image and how information flows through these connections that are degrading.

Multiple sclerosis, in particular, is a disease in which the patient doesn't know for certain what their evolution will be like, it's highly unpredictable...

 

FP: It's one of the saddest diseases there is. It can occur in young people, some 70% are women and 30% men. Of the four types of multiple sclerosis, there's one group of patients who might have an attack one day and then it'll be another 20 years before the next one. Others might have an attack one day, another one a couple of months later, and within a year they're in a wheelchair. It's a terrible disease for the patient, and it's also very tough on the family. We don't know exactly what the triggers are, we scientists can't get a hold on it, and there's still a lot of research to be done both in terms of diagnosing and treating the disease.

Today, what areas of health are harnessing artificial intelligence the most?

 

JCR: In terms of research, a lot of things are being done in areas such as oncology, multiple sclerosis, genomics... What I'm not so sure about is to what extent the research has moved over into clinical practice, in other words, how it's benefiting the patient.

FP: Artificial Intelligence in research is everywhere, but in clinical practice, it's difficult to find cases where it's extensively used. It's being tried, but the uncertainties and the black box that artificial intelligence is for doctors means that they're rather reticent for it to become part of the everyday procedure with patients, at the health centre or in hospitals. We're at a moment of transition.

Will the time come when artificial intelligence has found its way into the day-to-day practice of check-ups in health centres or hospitals?

FP: I think it will. The methods are increasingly more precise and robust, and it's more widely accepted in areas such as medical imaging, diagnostics and epidemiology. To come back to the COVID-19 project that we're working on, we're proposing an algorithm but I suppose there'll be doctors who'll be resistant to it.

Why would they be resistant to it?

FP: Because of the uncertainty, the fact it can fail. An artificial intelligence-based method can fail by 10-20%, a percentage that doctors are uncomfortable with, as of course are the patients. Although in some cases, the method can be more precise than some experts, we're at a point where AI has to find its place in everyday real clinical practice. Do we want AI to be a substitute or an aid? If it's an aid, it has a lot of scope, but if we want to substitute a doctor for it, we evidently need to improve the algorithms a lot to get this point of human touch.

JCR: I'd give an analogy using autonomous cars. It'll still be a long time before we see cars driven autonomously on the roads, but that doesn't mean that the cars sold don't have aids: when you go near the white line or you get too close to another vehicle, you get a warning, the steering is corrected if you slide in a curve... In the case of artificial intelligence applied to health, it's a bit like that. In this case, though, legal aspects are involved as well. For example, if the car has an accident, is it the fault of the driver or the manufacturer? Doctors might be reticent because they may find themselves facing a similar problem: if the diagnosis is wrong and the patient sues, who's at fault? The doctor or the machine that took the decision?

They're complex issues and all the pieces have to fit together: technological, social and legal ones. What I do believe is that the path has already been laid out and there's no going back. It'll happen sooner or later, but we'll see it increasingly more integrated and included in aid for doctors.

FP: Yes, it's an impending force that will become a part of the day-to-day doctor-patient dynamic, come what may. What we have to find is how it fits in.

Will artificial intelligence end up diagnosing in the future or is this a role that will always lie with the doctor? In the future, will we go and see a robot doctor that will tell us what we've got using AI? Is this a plausible scenario?

JCR: It's perfectly possible. But it's another matter if you as an individual have confidence in a machine or not. And legally, who will cover you if the machine gets it wrong? The possibility of doing it with a low error is there, although that doesn't mean that a human can't take a look at it and assess it. What I think is that at a first level of care (to give a simile, at a level similar to calling the emergency services), it would be possible to be attended by an AI algorithm which, based on your description and symptoms, is able to give you advice for a simple treatment or refer you to the right professional or health centre.

What are the red lines of artificial intelligence in terms of health? Will it be our privacy that pays the price?

JCR: Ethics, and we'd include privacy here, is the main one. In China, there are millions of CCTV cameras on the street that monitor where people go, and in these times of pandemic, it's easy to monitor people's mobility. But this brings with it significant problems regarding privacy and ethics, which we in Europe are not willing to give up. We have to fight to maintain our level of privacy, and we have to ensure that AI has an ethical component and is fair with people and that there is no discrimination on grounds of race, gender or any other.

There are currently Catalan, Spanish and European groups who are putting forward route maps about the direction AI should take so that it becomes an asset to society and not an exhaustive control of the population: from how it should fit into society, how we need to take responsibility for it, who should be responsible for certain things, what the limits are... It's a sector with a lot of potential, both in terms of the benefits it offers society and from the economic and job-creation point of view. But it has to be rolled out carefully so that it doesn't end up like in the film 1984, based on the novel by George Orwell, where the idea of Big Brother comes from.

FP: Every AI-based project goes through an ethics committee, which means you process data seriously. You don't ask for random data, the data aren't kept for longer than is strictly necessary and you avoid asking for more data than is called for. You have to think what it is you want to do and what data you need so as to place limits on it all.

As well as working on medical projects, your group also has a social project linked to energy poverty. What does it consist of?

JCR: The project arose from a meeting with the coordinator of the iSocial Foundation, a body that works on innovation in the social sector. They wanted to see if you could use artificial intelligence algorithms to try and predict how energy poverty might evolve in specific neighbourhoods and to predict the probabilities that a family or household has of not being able to pay their electricity or water bills. At the same time, we got in touch with Santa Coloma de Gramenet Town Council, which is a town where there's a high proportion of families living with energy poverty. Once we secure funding and we can start researching, we'll need to gather very diverse data (economic, income, late payments, bills, energy rating of homes, family expenditure, loans owed, employment status, dependency problems, etc.) which should enable us to analyse the level of energy poverty and its evolution in the neighbourhoods, as well as to predict which households may have high probabilities of being unable to pay their bills within a range of a few months to a year.

Normally, when a town council detects a situation of energy poverty, the family's been in it for months and has stopped paying bills or is being threatened with being cut off. The problem has got big. Essentially, our aim with the project is to help Social Services in the early detection of possible cases of energy poverty.