Meet the women using AI to detect hidden heart disease

Australian startup Sonorus is developing an AI algorithm that can identify signs of rheumatic heart disease (RHD) within minutes, simply by analysing heart sounds. Showcased at the recent SXSW Sydney Tech and Innovation Showcase, Sonorus' AI is trained on cardiovascular audio to pick up signs of heart disease that might otherwise go undetected.

"It's a low cost, portable mass screening and triaging tool, where, rather than waiting for someone to develop signs that they have something wrong with their heart, we use it as a preventative check for susceptible communities that may be at risk of it," Sonorus CTO Leah Martínez told Mashable. "We screen them early, catch it early, and get them the treatment they need before they even need to potentially think about getting an echocardiogram."

The goal is to make RHD easily detectable via a simple, routine check — one that's quick, accessible, affordable, and may not even require a doctor.

Starting Sonorus

Sonorus began as an idea from Martínez's friend and co-founder, CEO Dr. Julie Dao, in 2022. The two had been in university at the time, with Dao completing her PhD in cardiovascular health while Martínez studied engineering. Having previously worked on a separate project together, Dao approached Martínez to contribute her technical expertise to her new "crazy idea."

"Her background is that she's from Vietnam, and so rheumatic heart disease is something that's really rampant over there," said Martínez. "It was just the two of us inside my two bedroom apartment, in the study, just figuring out how to make a circuit work…. When we started this project at that time, I was in the middle of my engineering degree, and I was still learning and picking up new things. And we built a very, very simple proof of concept, just enough to say that you can record heart sounds with some simple, low cost technology."

Martínez is now nearing the end of her engineering degree, as well as holding down a day job and working at Sonorus. Boosted by accelerator programs from supporters such as Monash University, Sonorus now has a fully working prototype, and is taking steps to collect further heart audio data internationally as it continues to develop its AI algorithm.

"We want to develop the world's largest data set of clinically viable heart sounds," said Martínez. "We're going beyond that idea that heart sounds are just what we are able to hear and perceive as humans, and actually look at it from a [perspective of], What more new information can we extract that no one's looked at before?"

What is rheumatic heart disease?

RHD is permanent damage to the heart valves caused by inflammation, which can cause blood to flow in the wrong direction. The disease starts from a Strep A infection, which develops into acute rheumatic fever (ARF), according to John Hopkins Medicine. This fever then causes inflammation which damages the heart valves.

Affecting around 55 million people worldwide, RHD kills approximately 360,000 people every year, according to the World Health Organization. It is most prevalent in lower socioeconomic communities where healthcare is more difficult to access, disproportionately impacting Indigenous populations in the U.S. and Australia, as well as people in countries across Africa, the Middle East, South-East Asia, and Oceania.

RHD currently can't be completely cured, however it can be managed with shots of penicillin every three to four weeks. Unfortunately, patients are unlikely to seek a diagnosis until after they're already experiencing symptoms, which may include chest pain, fatigue, and shortness of breath. Left untreated, RHD can significantly impact a patient's quality of life, increase their risk of stroke, and lead to heart failure.

As such, Sonorus hopes its AI will catch RHD before major damage to the valve occurs, allowing swift intervention to prevent further degradation. Early detection can significantly reduce the treatment required, costs involved, and impact on the patient.

"The risk [posed by RHD] grows exponentially the longer you leave it," said Martínez. "You can treat it with a $5 shot of penicillin [if you catch it early enough], but if you catch it too late, you're looking at open heart surgery."

Using AI to diagnose heart disease

Echocardiograms are the current gold standard for detecting and diagnosing RHD, using ultrasound to create images of the heart that doctors can examine. However, while echocardiograms are an undeniably useful diagnostic tool, they do have limitations.

By the time a patient is sent for an echocardiogram, issues with their heart are typically already apparent, with damage severe enough that a doctor can detect it with a stethoscope. Sonorus aims to detect RHD before it gets to that point, utilising AI to identify characteristics in heart audio that may indicate issues but aren't identifiable to the human ear.

"[If you do] an ultrasound, you can see physically where the damage is," Martínez explained. "But what we found is there's a couple of precursory markers before you even get to that point, just by listening to the heart and listening to how it operates after you've caught that disease."

________________________________________________________________________________________________________

A new Mashable series, AI + Health, will examine how artificial intelligence is changing the medical and health landscape. We'll explore how to keep your health data safe, using AI to decipher your blood work, prompting chatbots effectively when it comes to health questions, and much more. ________________________________________________________________________________________________________

Echocardiograms are also expensive, time-consuming tests which require specialist training to administer, making them unideal for screenings of large groups of people. Sonorus' founders claim that their portable device can carry out such screenings instead, helping to identify patients most in need of further examination.

"When you're thinking about the cost of an ultrasound machine just by itself, it's about maybe 10 to 20 grand, then you have to think about the cost of an operator, cardiologist, technicians, admin staff, that sort of thing," said Martínez. "And it just runs you up really fast."

Ensuring that Sonorus' devices are cost effective is one of the company's primary aspirations. While they're still a long way from going to market, Martínez speculates that their devices may ultimately cost under $AU1000 (around $700), or perhaps even half this amount.

"We want it so that it's so affordable that it can be scaled and deployed out as fast as possible to so many different people," said Martínez.

Sonorus' AI aims to assist doctors, not replace them

Sonorus isn't the only organisation working to detect heart disease via AI sound analysis. Studies on AI-enabled stethoscopes have found that such tools could significantly improve doctors' ability to detect heart issues, more than doubling identification of some conditions when compared to analogue tools. 

However, Sonorus distinguishes itself by being designed for use by the public rather than healthcare professionals. As such, ease of use is a high priority in Sonorus' development process to enable widespread deployment. Still, Sonorus is intended to be an assistive tool and first step toward diagnosis, not a replacement for a doctor's professional assessment.

"We're not creating a crutch for doctors and saying, 'this is the new way to diagnose it, and you have to take this as the Bible, and now you don't have to do any of your due diligence,'" said Martínez. "What we're actually doing instead is a [screening] process... so [doctors] can focus on the cases that they actually need to get to."

Challenges: Collecting data, consulting with communities

For Martínez, gathering the data necessary to train Sonorus' algorithm has been their biggest hurdle so far. While Sonorus did use open source data in a preliminary validation of its algorithm, Martínez explained that it wasn't of the quality they wanted. The company is now working to gather its own data, and hopes to have a database containing hundreds of thousands of data points within the next few years.

"Everyone that works in AI, they always say that your algorithm is only as good as your data is," said Martínez. "We're trying to collect data straight from the [the communities Sonorus is designed to serve], and making sure that it's high quality and what we need."

Sonorus is building its systems in consultation with leaders in the communities it seeks to serve, so that its targeted users are involved in the development process. By doing this, it aims to ensure that the types of solutions it offers are ones that the community will trust, find simple to operate, and want to use. For example, Martínez noted that autonomy regarding health is important to Indigenous Australian and Pacific Islander communities. Including them in Sonorus' development helps to address this need, building understanding and integration into the community.

"We don't want to just be like, 'here, we're going to design a solution, here you go, go ahead and use it,'" said Martínez. "Because we know a lot of the time that healthcare is partly about the solution, but it's also partly about the care in the relationship that you build with the people that are going to use it."

The future of cardiovascular diagnosis?

Martínez believes that the untapped potential for medical use is one of the biggest blind spots regarding audio and AI research right now.

"A lot of AI [concerning sound] right now is really used for audio processing, speech to text generation, but no one's really using it for clinical applications, and so we're filling in that gap," Martínez said.

Though Sonorus is currently developing its AI solely to detect RHD, Martínez believes the technology could be applied to other medical conditions as well. The hope is that, in the future, Sonorus will be able to screen for multiple different cardiovascular conditions with just one simple test.

"Honestly, that would be the goal," Martínez said. "We're starting with rheumatic heart disease, so we want to make sure that we get this one right, and make sure that the solution works. And then from there, because AI algorithms are so easily scalable and applicable to different things... we want to move to other valvular diseases as well."

________________________________________________________________________________________________________

The information contained in this article is for educational and informational purposes only and is not intended as health or medical advice. Always consult a physician or other qualified health propeovider regarding any questions you may have about a medical condition or health objectives.

Disclosure: Ziff Davis, Mashable’s parent company, previously filed a lawsuit against OpenAI, alleging it infringed Ziff Davis copyrights in training and operating its AI systems.