Imagine if your smart watch could monitor chronic health conditions such as heart disease and relay that information securely to your doctor? The idea that data captured by genomics, sensors, and wearable technologies could one day lead to affordable, patient-friendly disease monitoring and management systems is part of the grand challenge facing Dr Elena Daskalaki and other computer scientists at ANU.

Dr Daskalaki, who is a Fellow in Computer Science, is part of the big data group of the Our Health in Our Hands (OHIOH) grand challenge. The project draws together health experience experts, clinicians, and researchers across multiple disciplines, to develop new technologies to improve the diagnosis and treatment of chronic diseases, particularly diabetes and multiple sclerosis.

The big data team, led by Associate Professor Hanna Suominen, is developing machine/deep learning algorithms that have the capacity to map enormous amounts of data from multiple sources to deliver personalised health support systems.

“Because this is usually complex data, simple statistics don’t work very well and so you have to go to more advanced processing techniques. This is where machine learning starts to be very powerful,” said Dr Daskalaki.

Combining medicine and math

Born in Athens, Greece, Dr Daskalaki was drawn to both engineering and medicine as a young person. “When I was trying to choose what I wanted to do with my life I was torn between medicine and engineering. It was really hard for me to choose, but I found a way to combine. I really like the multi-disciplinary component of biomedical engineering,” she said.

Following her electrical and computer engineering studies at the National Technical University in Greece, Dr Daskalaki went on to do a PhD in Biomedical Engineering with a focus on the design of algorithms and prediction models for glucose regulation in type 1 diabetes; a focus that is a big part of her OHIOH work.

“I really like the collaborative, multi-disciplinary components. You feel like you’re doing something important for the world. It’s very motivational,” said Dr Daskalaki.

As part of the research delivered by the OHIOH project, Dr Daskalaki supervises a number of course-work and PhD students working on the design and development of machine/deep learning algorithms for application in healthcare.

Currently the team is looking, for example, at glucose prediction models and the development of an artificial pancreas system.

The artificial pancreas is a closed-loop control system that tracks glucose levels using a continuous glucose monitor (CGM), and automatically estimates the insulin required which is delivered via a pump.

It’s estimated that there are on average seven new cases of Type 1 diabetes diagnosed every day (Australian Institute of Health and Welfare). The artificial pancreas is an “all in one” diabetes management system, replacing the reliance on finger-stick testing and multiple daily injections.

The team is exploring how to map information from multiple sources such as CGM, insulin, food intake, heart rate, and other physiological data into deep complex representations to predict the future CGM profile in 30 minute, and 1- and 2-hour horizons, as well as to optimise insulin delivery for better and safer glucose regulation.

The team is also developing natural language processing algorithms to simplify medical texts to assist with better diabetes management in schools. The aim is to deploy deep learning to detect complex words, resolve acronyms and find the optimal substitution of simpler words or definitions so that people without a medical background, such as teachers, can understand the disease facts and be guided on what to do in certain circumstances, for example a student experiencing a hypoglycemic event.

A critical part of the big data work is to realise interpretable machine learning solutions that help identify which parts of the data are important for the diagnosis and prognosis of diseases. These studies include contributions not only to detecting and managing diabetes, but also classifying and predicting neurological conditions such as multiple sclerosis and Parkinson’s Disease. Dr Daskalaki believes that this is a crucial feature that is expected to boost significantly the acceptance of these tools in clinical practice.

Eyes: a window to the brain

According to Dr Daskalaki our eyes can tell us a lot about our health. “It’s often said that the eyes are the mirror of the soul: the eyes are the mirror of the brain,” she said. “Neuroscience shows that the response of the eye can reflect not only diseases of the eye but also other diseases, even cognitive diseases.”

Because our eyes are an important part of our larger nervous system and share a number of vascular and neural similarities to the brain, by measuring responses to stimuli in the eye you get an insight into brain pathology.

People with diseases such as multiple sclerosis and Alzheimer’s Disease also often report visual symptoms which may point to potential ocular biomarkers for these diseases.

Drawing on the relationship between the eyes and disease, Dr Daskalaki is looking at computational methods for rapid and objective eye and brain testing of a variety of ocular and neurological diseases based on the response of the pupil to certain stimuli.

Creating powerful diagnostic tools based on the response of the eye offers the promise of non-invasive disease monitoring. And it’s this potential that Daskalaki is exploring as part of an industrial translation project led by Professor Ted Maddess, from the John Curtin School of Medical Research (ANU), in collaboration with Konan Medical.

Professor Maddess and Konan Medical have developed a device that examines the health of the eyes, much like the standard visual-field tests that you might have when seeing your optometrist. However, unlike the standard visual-field tests where patients are required to press a response button hundreds of times, the Objective Field Analyzer sends stimuli to different regions of the eyes and records how they respond. Dr Daskalaki is working with Konan Medical and Maddess to augment the possibilities of the device.

“The idea is that we want to model the response of the eyes and use this model in order to associate it with different diseases,” said Daskalaki. “For example, a person who has glaucoma has a different pupil response to a stimulus pattern compared to someone who doesn’t. But the question is what is the difference?”

“Obviously a healthy eye responds differently but you have to find the right way to stimulate it so that you get a response that differentiates between the two eyes. You need the right feature in your data to be able to distinguish between the two things.

“Identifying where the difference lies and associating it with a particular biomarker is where machine learning has significant applications.”

So what of the future of smart technologies and health care? Daskalaki believes that the world is moving to home-based and much more personalised solutions. “Smart watches are already measuring a range of medical signals, your heart rate, your oxygen saturation. Many people are wearing them but they’re not yet linked to your doctor,” she said.

While Daskalaki acknowledges the challenges around data privacy, she thinks that at some point there will be a bridge between the two allowing people to self-monitor at home while keeping up with regular visits to health care professionals.

Perhaps one day the smart watch will become more than a gadget that allows us to take calls and track our heart rate? It could become part of a highly personalised solution to manage chronic disease.