AI Stethoscopes: Unlocking a Billion-Dollar Heart Health Market

The stethoscope is an icon, but let’s be honest, auscultation has its limits. Even for a seasoned clinician, that familiar thrum against the chest can be misleading. It’s especially tricky with structural heart disease, where the first sign is often a subtle murmur that’s incredibly easy to miss during a routine physical. Now, AI-enabled auscultation is changing the game, giving clinicians a way to hear what the unaided ear might otherwise ignore.

The Unseen Burden: Valvular Heart Disease and Missed Diagnoses

More than 8 million Americans have valvular heart disease (VHD), a serious public health issue that can have grim outcomes if it isn’t caught and treated. Detection usually starts with a stethoscope, with a clinician listening for those tell-tale murmurs. But how well does that actually work? The truth is, the sensitivity and specificity of manual auscultation are all over the place, getting thrown off by everything from the doctor’s own experience level to a noisy room or the patient’s body type. Eko Health, a company that’s been pushing digital stethoscope tech, points out in its FDA clearance documents that traditional exams miss up to 57% of clinically significant VHD cases. That’s a staggering number of missed diagnoses, and it means we desperately need better tools right there at the point of care.

Eko’s Murmur Analysis Software: Augmenting the Human Ear

Eko Health’s Murmur Analysis Software (EMAS) was a direct attempt to plug that diagnostic gap. The software, which works with Eko’s digital stethoscopes, uses AI to interpret heart sounds for the clinician. EMAS first got its 510(k) clearance from the FDA back in July 2022 FDA 510(k) clearance database for Eko EMAS. At the time, Eko’s own data claimed an overall sensitivity of 85.6% and specificity of 84.4% for spotting murmurs. For adults over 18, the numbers were even better: 90.2% sensitivity and 90.6% specificity. Independent research backed this up. A study from Massachusetts General Hospital found that Eko’s algorithm literally doubled the rate at which structural murmurs were identified in primary care compared to standard practice FierceHealthcare article on MGH study. Putting this tool in a primary care workflow could drastically improve early VHD detection, ensuring patients get referred to specialists much sooner. Eko didn’t stop with murmurs. They also worked with the Mayo Clinic on an algorithm to spot low ejection fraction (low-EF), a key sign of heart failure which was later validated in a study by Imperial College London, proving the technology could be expanded to detect more than just structural problems Imperial College London study on low-EF detection algorithm.

The Evolution to EFAST: A Foundation Model Approach

The world of cardiac AI is moving fast, and the focus has shifted from single-purpose algorithms to more complete “foundation model” approaches. In September 2025, Eko Health announced it had received FDA clearance for its EFAST algorithm, which they called the first FDA-cleared foundation model for cardiovascular AI Eko Health announcement on EFAST clearance. This was a completely different beast. Instead of being trained to do just one thing, EFAST was built on a massive dataset of over 4 million de-identified heart sound and ECG recordings. By learning from such a huge pool of data, the model can recognize much more complex patterns and relationships between different cardiac signals. EFAST was cleared to detect both structural murmurs and AFib. Importantly, the clearance announcement noted improved specificity over older algorithms, which is essential for reducing false positives that lead to unnecessary patient anxiety and costly downstream tests. Building foundation models like EFAST is a huge step toward creating a single, versatile AI tool that can interpret a wide range of cardiovascular data, almost like having a cardiologist’s expertise built into the device. Using a giant dataset like this gives the algorithm a more nuanced and accurate way to interpret auscultation, which may also help prevent problems like algorithmic drift by constantly learning from real-world data MedTech Innovator article on Eko EFAST.

Distinguishing Foundation Models from Single-Task Algorithms

So what’s the real difference between a foundation model like EFAST and a simpler single-task algorithm? It’s key to understanding where cardiac AI is headed. A single-task algorithm is trained to do one job, like finding a specific murmur, and it’s optimized for only that job. It might not work well if it encounters a different condition or even slight variations in the data. A foundation model, on the other hand, is built from the ground up to be flexible. By training on a gigantic, diverse dataset of raw heart sounds and ECGs, EFAST develops a deep, generalized understanding of heart physiology. This lets it do several related jobs, like finding murmurs and AFib, and it can likely be adapted to learn new tasks with much less effort. The architectural change provides some big wins:

• Generalizability: These models handle variations in patients and clinical scenarios better. They’re less likely to be thrown off by an unusual case because their vast training data has exposed them to so many different presentations.
• Efficiency: It’s incredibly resource-intensive to build and validate a separate AI for every single cardiac problem. A foundation model offers a more direct path to expanding what a device can do without starting from scratch every time.
• Robustness: The sheer volume of training data makes the algorithm tougher. It’s better at ignoring a noisy background and picking up faint signals that a narrowly-trained AI would probably miss.

This shift from specialized AI tools to broader foundation models shows that AI in medicine is maturing, much like we’ve seen with large language models. For cardiac screening at the point of care, it means we’re getting closer to a single device that can help a clinician screen for a whole range of heart conditions with much better accuracy.

The Clinical Impact and Regulatory Pathway

The practical impact of AI-enabled auscultation is most obvious in primary care clinics and other non-specialist settings where you don’t typically have a cardiologist on hand. These devices provide an objective, AI-driven analysis of heart sounds, essentially acting as an expert screening tool that can flag a suspicious finding for a specialist’s review. This can get patients with VHD and AFib diagnosed and treated earlier, which improves their outcomes and reduces the long-term burden of their disease. Both EMAS and EFAST used the FDA’s 510(k) clearance pathway, which is the standard route for medical devices that can show they are substantially equivalent to (and an improvement on) a device that’s already on the market FDA guidance on 510(k) submissions. This path allows for faster market access for new technologies that build on established device types. Eko’s journey from EMAS to EFAST shows a smart, iterative approach to improving their technology within that regulatory system. The fact they could demonstrate better specificity with EFAST shows just how rigorous the validation has to be for these AI tools. The AI stethoscope, as shown by what Eko Health has been doing, is fundamentally changing what’s possible in point-of-care cardiac screening. By adding a layer of smart AI to the traditional physical exam, these devices are already helping catch serious heart conditions that would have gone unnoticed. The move to foundation models like EFAST just promises to expand that capability and accuracy, making early detection of cardiovascular disease a more achievable goal in everyday clinical practice.