A diagnostic tool for respiratory disease:
machine learning coughs up
ResApp uses machine-learning algorithms to analyse cough sounds and diagnose conditions such as asthma, pneumonia and bronchiolitis from a smartphone. Natalie Healey speaks to creator Professor Udantha Abeyratne to find out how the platform could help patients worldwide.
t’s likely that coughing ranks up there with your least favourite sounds. A fellow train passenger’s persistent hacking can feel impossible to tune out, for instance. And it’s even worse for the cougher themselves, who has to contend with feeling rotten while worrying about how the noise is affecting those around them. But as Udantha Abeyratne points out, we shouldn’t be humiliated by a noisy cough — not least as we can’t help it — because it can give doctors important clues about respiratory disease.
“Cough is the body's natural reflex mechanism to clear the respiratory system of foreign materials and various internal secretions and obstructions,” explains the associate professor at the School of Information Technology and Electrical Engineering at the University of Queensland and creator of the technology behind smartphone app ResApp — which analyses the sound of a cough to determine which disease is causing it.
Doctors already diagnose some conditions based on the type of cough they produce. Croup can be identified largely by the ‘barking-ness’ of the accompanying rasp. While whooping cough also has a distinctive ‘whooping’ sound — hence the name. However, these noises can still be misinterpreted by experienced clinicians. Other respiratory diseases are even more difficult to determine. Could it be a simple cold or something more serious, and should you wait for it to go away on its own or risk an unnecessary visit to the doctor?
“When we consider the complex nature of coughs in other diseases, it is not feasible to describe them in words, and it’s more difficult to train a human to identify them accurately,” points out Abeyratne.
The problem with pneumonia
In 2009, when the Bill and Melinda Gates Foundation called for proposals for tools to help diagnose pneumonia in resource-poor regions of the world, Abeyratne had an idea. He had previously found that snoring sounds could help diagnose sleep apnoea and believed he might be able to do something similar with coughs for pneumonia.
Pneumonia can be serious even in developed countries like the UK and Australia, but it’s an even bigger problem in more remote areas of the world. Approximately 1.4 million children under the age of five die from the disease worldwide each year, according to the World Health Organization.
The majority do not have access to diagnostic facilities, such as X-ray imaging and lung function testing. If a simple tool could be developed, parents could use it to determine whether urgent medical attention was required for their child.
“The foundation awarded the grant, publicly praising my idea as an innovative solution to a tough problem. This was the beginning of ‘Project ResApp’ at the University of Queensland,” reveals Abeyratne.
Artificial intelligence keeps learning
But how can sound work as a diagnostic tool? Abeyratne explains that during coughing, our lungs connect to the atmosphere via a column of air. This column supports a much higher bandwidth across the chest than normal and can generate sound frequencies up to 60kHz — far beyond the human hearing range.
“In pneumonia, large sections of contiguous lung tissue can get infected, resulting in an accumulation of secretions in bronchial areas and in the tiny elastic air sacs (alveoli), which limit the total airflow volume/rate, as well as propagate particular sound characteristics,” he says.
Although the human ear wouldn’t be able to distinguish these sounds as signs of pneumonia, a machine could be trained to spot them. Abeyratne has worked on machine-learning projects for much of his career. This artificial intelligence (AI) technique constructs algorithms which are constantly tweaked as the device learns new information.
He and his team started by matching signatures in a large database of coughing sound recordings with known clinical diagnoses (such as pneumonia, asthma, bronchiolitis and chronic obstructive pulmonary disease). Their AI tool was then developed to determine the most likely condition when a new cough sound is detected. As the platform only requires sound, rather than physical contact with a clinician, a smartphone microphone could be good enough to deliver an accurate diagnosis.
“Our work indicates that we do not need to rely on spontaneous coughs. Voluntary coughs are sufficient. The process has been fully-automated and implemented on an iPhone,” says Abeyratne.
Commercialising the technology
In 2015, the research commercialisation arm of the University of Queensland, called UniQuest, licensed the cough-centred technology to research spinoff company ResApp Health. The aim was to be the world’s first diagnostic app for respiratory disease — one that patients could use to get a diagnosis without even leaving their house.
The team first had to prove its platform could produce a comparable diagnosis to a doctor. In clinical studies, the algorithm was tested against a team of clinical adjudicators using state-of-the-art technology.
“The reference clinical diagnoses had access to resources including results of diagnostic tests that are not normally available to a frontline clinician at the time of a routine consultation in the real world. Our app, however, provided an instantaneous result based on the in-situ measurement of cough,” Abeyratne reveals.
ResApp listed on the Australian Stock Exchange in July 2015. And in August 2019, following a clinical trial published in the journal Respiratory Research, ResApp received CE mark classification for both adults and children, meaning it can now be sold in Europe.
Abeyratne has hopes that ResApp will assist humanitarian organisations in the developing world in future and will continue actively working with the company. But he is also busy with separate research and development projects at the university that may one day be commercialised too.
“I’ve always believed in the concept of ‘the full cycle of research’, starting from creative ideation down to research translation and commercialisation. I would like to continue along that path into the foreseeable future,” he says.
Share this article