Med-Tech Innovation News spoke to Iulian Circo and Joe Brew, co-founders of Hyfe, about its continuous monitoring technology for lung health.
First of all, tell us about Hyfe. How did you get into the medtech space?
Hyfe was founded in early March 2020, in response to the emerging COVID-19 pandemic. The team - which is global and includes top-tier academia, engineering and entrepreneurship - came together animated by the idea of building reliable A.I. tools that can detect cough and monitor respiratory conditions. We wanted this tech to be deployed at scale via smartphones and wearables.
Tell us what inspired the idea?
Cough is everywhere, yet somehow no one has a reliable way to quantify coughs. Not patients, not doctors, not researchers. Even more, we don’t even have a vocabulary to discuss cough in a meaningful way. Think about it: How many times did you cough yesterday? You probably can’t remember. It's normal for a healthy person to cough about a dozen times each day. Now try to describe those coughs. Maybe you will say it was dry? Perhaps it was, but in using that word we have reduced an incredibly rich signal to a single word. In other words, we’ve been throwing away important data on an important symptom (cough).
Even in in-patient settings, where a patient is surrounded by instruments that quantify every possible biomarker, cough continues to be evaluated based on the same simple, qualitative exchange. The physician doing morning rounds asking, “How has your coughing been?” and the patient providing a vague, information-poor answer based on their own perception which frequently fails to correlate with an objective measure. We would never treat other health data signals with such a lack of rigor - can you imagine a doctor asking a patient, “How’s your arrhythmia?” They don’t ask, they measure. So why then, with cough, do we ask instead of measure?
Hyfe technology provides a reliable instrument that counts cough frequency over time, providing medical professionals with accurate, granular, quantitative information that fills a gap currently filled with vague impressionist, biased, incomplete, inaccurate estimates. This can help them gauge the state of a patient as well as the ability to monitor cough over time and measure correlations with treatment decisions.
Give us some insight as to the technology behind it, how was it developed?
Hyfe uses deep learning techniques – namely Convolutional Neural Networks, that learn by processing large quantities of data. We understood from the beginning the importance of quality, real-world data for a machine learning platform. This is why we launched a consumer product early on and ensured that we continuously increase the value we create for regular folks - mostly chronic coughers - without charging a cent for the technology. This allows us to process millions of real-world coughs and cough-like sounds, which is why our detection models are in a league of their own in terms of performance. We also partnered with dozens of globally respected research institutions and provided them with reliable cough detection instruments as part of clinical trials. This gave us access to high quality ground-truth data which we are using to train more sophisticated ML models, teaching them what specific diseases “sound like.” Imagine being able to detect a respiratory condition - say lung cancer - before symptoms become noticeable.
In an interesting real-world story, one of our users saw an increase in their cough frequency on their chart, which they were not aware of. They subsequently tested for COVID-19 and found that they were positive. This person would have been an “asymptomatic” case because people are really bad at understanding subtle changes in their cough frequency.
How can the technology help differentiate between the different respiratory diseases you help to diagnose?
Doctors have been listening to coughs and making clinical decisions for generations. In fact, the first telemedicine consultation ever (Reported by The Lancet in 1879!) was a doctor listening to a child’s cough via phone ('Lift the child to the telephone, and let me hear it cough... That's not the croup’).
And in clinical practice, it’s very common for doctors to ask about, observe, and listen to coughs, and make decisions based on that information. This is why we know that coughs carry a lot of information. And we know that machines are much better at identifying subtle unique signatures than the human ear. We are also uncovering more and more evidence that there is information not only in the sound structure of an individual cough, but also in the frequency of coughing over time - clusters of coughs after meals, for example, or at night, or during outside activities, or when a person is supine, etc. Combining these different signals will help practitioners diagnose respiratory conditions, in particular in low infrastructure settings where accessing a laboratory may be more complicated/ expensive.
We already have achieved promising early results with diseases such as TB or COVID-19. Of course, bringing diagnostic tools on the market requires much more than technology alone - and we are now in the process of gathering scientific, objective evidence for some of these tools that will help us get the regulatory approvals required to bring Acoustic AI diagnostic tools to the market.
Your product seems to be a case of trend spotting to get a fuller picture, do you think this is part of a trend towards more preventative treatments?
Of course. This is a very important trend. As we learn more about effective ways to prevent disease - and therefore decrease negative outcomes as well as costs - we will see a big shift towards more preventative and personalised treatments. We will see insurers and payers incentivising providers to promote preventative measures, we will see more tools put in the hands of patients and we will see a lot of innovation.
Additionally, there are a few other important and related trends. Smartphones are now ubiquitous and they are redefining a whole generation’s interaction with the world - from dating to banking to experiencing the world. Think about it: nearly everyone is now carrying around a computer with a microphone in their pocket. This is a huge opportunity for those who know what to do with sound. Healthcare will be impacted by this trend - as it should be. For us, every smartphone is a powerful piece of infrastructure - rich in sensors and always close - that can become a powerful ally in bringing healthcare into the modern world. And smartphones also present a huge opportunity for development and inclusion, bringing healthcare and diagnostic services to people for whom a doctor is unreachable. With the increasing access to internet connection and the development of cost-effective health technologies as Hyfe, we believe that the future of healthcare - preventative and essential - is digital and accessible through daily devices, such as smartphones, smart wearables and potentially implantable sensors.
What effects did COVID-19 have on you as a company, and what do you think it has on the space you operate in?
COVID-19 has marked the end of an era and the beginning of a new one. In the process, it has accelerated some much-needed change - around the virtualisation of services and the integration of technology in processes that have stubbornly refused to change for decades. It has also woken humanity up from a slumber in which the “era of infectious disease” seemed like something from the past.
As a company building cough detection tools during a respiratory pandemic, we were definitely ahead of a curve. But we don’t see Hyfe as a “COVID project”. On the contrary, COVID will come and (hopefully) go, but other respiratory disease pandemics - like TB - are unfortunately here to stay for quite a while longer. Non-communicable diseases resulting in chronic cough will stay, too. We think that acoustic epidemiology generally - and Hyfe specifically - will play a pivotal role in improving health for people around the world.
What obstacles remain when it comes to adopting technologies like yours?
There are several categories of obstacles that are worth mentioning. One of them is regulatory. As always, innovation and technology evolve much faster than the regulatory framework. So, we need to find ways to make a pure software-based platform using acoustic AI fit regulation that has been designed for things like scalpels and physical instruments.
We see that people are now getting familiar with machine-based predictions and decisions for health, yet research shows that they trust AI until a mistake is made. Therefore, we have an experienced multidisciplinary team working to ensure the quality and reliability of Hyfe algorithms.
There are also important privacy issues that we have to address. Sadly, we live in a world where people’s trust has been abused by companies who mismanaged personal data. In this environment, we are keenly aware that asking people to allow their smartphones to listen to their coughs is a pretty big ask. This is why for us privacy and security are paramount.
Anything else that you would like to add?
There is this wonderful quote from Freda Lewis-Hall: We have heath from Star Wars and healthcare from the Flintstones” and indeed, even while all aspects of our lives have been digitised and improved with help from technology, healthcare - the interface to health - has remained in the era of telephone bookings, five minutes in-person visits, and receipts sent by fax machine.
It is exciting to see how digital health evolves and adapts to changes in people's lifestyles. With the growing numbers of people spending time online, the ideas of virtual worlds and living in a “Metaverse,” healthcare and objective health monitoring could change drastically, yet we keep our minds open and constantly look for new opportunities, collaborations and solutions in improving health.
