Artificial intelligence (AI) and smart wearables will bring unprecedented advances in medicine. In fact, AI is already helping to make measuring devices in hospitals smarter. For example, these devices can automatically detect epileptic seizures using a cardio-encephalogram (EEG) or identify tumor tissue in images. The algorithms are also very versatile and able to learn new things easily (such as adding a new type of measurement or reading). One important area in which AI is breaking through fast, is radiology. This is because algorithms can handle images easily and learn from the notes made by radiologists.
And thanks to wearables, taking various readings from patients in hospitals can be made more comfortable for them—plus patients can also be monitored at home. The best example of this is in sleep research. At the moment, if you have a sleep disorder, you may have to spend a day in the sleep lab, which is an expensive exercise. But imagine if, instead, you could wear sensors for a month while sleeping in your own bed. Even if these sensors are not as accurate as the ones used in hospitals today, the results would in fact be more reliable because you will be producing readings in a realistic environment and over a longer period of time.
Another good example is monitoring premature babies in a neonatal intensive care unit. For these little tots, not only is it more comfortable to be monitored with wearables (or sensors in the mattress), but it can also be a matter of life or death. Working with the neonatology department at Leuven UZ (University Hospital), BioMED is researching the causes of stress in these babies.
Interestingly, wearables will certainly not always take the form of a device that you wear on your body. For instance, sensors can work just as well if they are incorporated into your bed or the seat in your car. When you order a new car, for example, you will be able to add on the ‘health check’ option—just as you do today when you go for heated seats. The same thing with your smartphone. In the old days, these phones were just used for making calls, but now they can do so much more. And in the same way, your car will become more than just a way of getting from A to B!
But it will take more than just good hardware and software to make this future come true. Researchers already see that it is no easy matter to convince doctors to work on research into the use of wearables for certain diseases. Why? First and foremost because wearables have to be reliable—and not ‘gadgets’. This means that certification will be very important for making a real breakthrough in this area.
It is also a fact that doctors simply don’t have the time to delve into the mass of data and specific interface of each wearable. Which is why it is important to supplement their training with more technical subjects and also expand their knowledge of AI. For that to happen, though, there will need to be a specialized team of biomedical engineers on hand to assist doctors to deal with this new source of information. After all, a doctor needs to be able to work with the patient and not be weighed down by computer work and huge amounts of data.
So, if we want to be able to use wearables as soon as possible in the day-to-day practices of GPs and specialists, some things have to change. Such as the reimbursement from the patient’s medical fund and the business model to be used. For example, will we buy the wearable from the pharmacy and then throw it away after use? Or is it something we can hire from our health fund? Or will we be given the wearable at the hospital and return it there after use so that a team can disinfect it and reboot it for the next patient?
So much for our vision of the future. But where do we stand today? And, more specifically, which of BioMED’s current developments give us a foretaste of how we will use wearables and AI for our health?
Picking up a pen with a sensor
Rheumatism is a common disease whose developments are monitored closely. At the moment, this is done by having the patient do standard exercises at regular intervals with a physical therapist, who fills in a score sheet, stopwatch in hand.
What if you no longer had to do these exercises, but instead wore a wearable round your upper arm that would recognize when you are putting your socks on, going upstairs, picking something up – and so on.
After all, these are also the actions that are simulated in exercises with the physical therapist. Working with BioMED, the rheumatology department at UZ Leuven has developed a system that consists of a sensor bracelet with an accelerometer which, thanks to BioMED’s ingenious mathematical models, recognizes specific actions in the day-to-day life of the patient. It also measures the time it takes the patient to do something, translating this to a score, just as they do at the moment at the hospital.
The wearable is now being expanded to include heart activity measurements (ECG) so that actions can be recognized even better. The advantage for the patient is that he/she doesn’t have to go to the hospital so often. While for the doctor the benefit is that the exercises reflect real life better and not just a specific moment in the doctor’s office when the patient is making a special effort. Many other measuring points can also be recorded over a longer time.
Mathematics helping to remove a tumor
It doesn’t always have to be wearables, though. The same mathematical models that make wearables smarter can also be used to give doctors more information about doing their work correctly. Example: removing a tumor. At the moment, MRI images are produced using a Magnetic Resonance (MR) scanner. The doctor then uses these images to identify the tumor tissue that needs to be cut out.
The composition of tumor tissue and healthy tissue is different, which can be seen in the spectrum of a piece of tissue, measured in the same scanner using MR Spectroscopy. When every piece of an MRI image is measured, this is called MRSI (magnetic resonance spectroscopic imaging). If the spectrum of the underlying tissue can be identified for each piece of the MRI image using mathematical models, the doctor is better able to distinguish the healthy tissue from malignant tissue, and it is even possible to see gradation differences within the malignancy inside the tumor.
And, because only a specialized spectroscopist can make sense of the spectral images, BioMED has converted the spectral information into ‘fingerprints’ that correspond with the recognizable tissue, while each tissue type is colored on the MRI image. This makes the tumor and its surroundings more straightforward to interpret for radiologists. These algorithms are now ready to be converted into a product that can be used by hospitals and which make MRSI easier to use for radiologists. Working with the radiology department at UZ Leuven and Ghent, the software has already been tested on identifying brain tumors.
Cutting out epilepsy
Some epilepsy patients find medication does not work for them. But, if the seizures are localized (focal epilepsy), then these patients can be helped by the surgical removal of that specific area of the brain.
Typically, patients spend a week in hospital for pre-surgical assessment, during which time they undergo all sorts of tests to localize the particular area of the brain causing the seizures as clearly as possible. EEG readings are taken using electrodes placed on the head, which measure the electrical potential differences (generated by the brain) in high time resolution. Functional MRI (fMRI) also provides important information, while high spatial resolution visualizes the areas of the brain that are active.
But what if you combined EEG and fMRI and took readings together in an MR scanner? Thanks to their complementary resolution properties, this EEG-fMRI combination makes it possible to localize epileptic activity better, both in time and space.
However, the flipside of the coin is that these EEG readings generate additional distortion in an MR scanner, which adversely affects the quality of the readings.
Working with the neurology department at UZ Leuven, BioMED has developed algorithms that enable the ‘epilepsy’ area of the brain to be better localized by combining EEG and fMRI to remove distortion. A software platform is now ready, but needs more multi-centric testing before the software becomes clinically usable.
A ‘hearing aid’ for epilepsy
To help bring a solution to the patient, it is of course always a good thing to have a company on board from the outset of a research project. This was the case in the highly successful imec.icon project, SeizeIT. Working with three companies—UCB, Byteflies and Pilipili, as well as the neurology and pediatric neurology departments at UZ Leuven—BioMED developed a wearable for epilepsy patients. This wearable looks like a hearing aid that is worn behind both ears. By combining the measurements of brain, heart and movement activity, epileptic fits can be detected.
In this instance, these are not the commonly known epileptic seizures in which the arms and legs make uncontrollable movements, but a type of epilepsy where the seizures are much more difficult to identify. To be able to monitor the patient properly and to know whether certain types of medication are working, it is very important to have objective and reliable information about the timing and number of seizures over a specific period. This is now possible with the SeizeIT prototype, which will be developed into a market-ready product as part of a major European follow-up project (supported by EIT Health).
About the author:
Sabine Van Huffel is Full Professor in Biomedical Data Processing at the Department of Electrical Engineering (ESAT) of the KU Leuven since 2002 and Program Director of the Master of Science in Biomedical Engineering KU Leuven. Her expertise techniques include EEG-fMRI, multiparametric MRI, ECG and EEG, focusing on epilepsy and neonatal monitoring. She received her Master’s degree in Computer Science Engineering, a postgraduate in Biomedical Engineering and a PhD in Electrical Engineering from the KU Leuven, in 1981, 1985 and 1987 respectively.