AI in cardiology: current developments and areas of application

Digitalization is rapidly changing cardiology—from continuous ECGs through wearables to AI-supported diagnosis of complex data sets. The potential of intelligent systems is particularly evident in cardiac medicine. They can detect patterns in ECGs, identify early warning signs of arrhythmia, or automatically analyze patient data.

“Telemedicine and digitalization can make more efficient use of healthcare resources and reduce costs in the long term. For example, early interventions make it easier to prevent secondary diseases and strokes, which reduces expensive hospital stays and acute treatment,” says the German Society of Cardiology. Wearables, smart implants, and telemedicine platforms open up new ways of monitoring and early detection—and at the same time reducing the burden on inpatient care.

One current example of this development is AI-supported stethoscopes, which were recently presented at a conference in Madrid. In this interview, cardiologist Dr. Philipp Breitbart explains where the practical benefits of this technology could lie. He categorizes the current study results, talks about the acceptance of digital devices in practice, and explains why AI only creates real added value if it is sensibly integrated into existing structures.

Studies on AI-supported stethoscopes were recently presented in Madrid. What were they about?

The question was whether an AI-supported stethoscope can be used to detect cardiac insufficiency, valvular heart disease, or cardiac arrhythmia more reliably. The studies, some of which involved more than 10,000 patients, have shown this: Depending on the condition, the device was able to make a diagnosis up to twice or three times as often as with a purely manual examination.

The stethoscope recorded an ECG at the same time, and the data was systematically analyzed. What was noticeable, however, was a relevant rate of false-positive findings, i.e., patients who were diagnosed with a disease that was later not confirmed. This is a fundamental issue in AI-supported medicine.

AI is strong at recognizing common patterns. In the peripheral areas—and this is where we very often find ourselves in medicine—it has weaknesses. It is precisely because diseases progress very differently from person to person that uncertainties arise. Patients who are actually healthy are then unsettled. On the other hand, the study showed that if a disease was actually present, the stethoscope was able to detect it with high sensitivity.

In the randomized comparative study, some practices received the AI stethoscope. The results showed that use was initially quite high but declined significantly over time. The possibility of feeding the results directly into the practice system was rated particularly positively—this saves work and increases acceptance. However, it should not be forgotten that the study has not yet undergone peer review. Only posters were presented at the conference.

Do you see such stethoscopes in practical use in everyday cardiology?

No. In cardiology, our standard diagnostics are ECG and echocardiography anyway. Before I even pick up the stethoscope, I've usually done the cardiac ultrasound, and that is much more meaningful.

I see a possible field of application in general practitioner care. GPs act as gatekeepers; they decide whether a cardiological examination is necessary. Such a device could be useful for them to obtain early indications. However, the high number of false-positive cases remains a problem, similar to smartwatches, which detect atrial fibrillation very sensitively but not always specifically.

Why could the technology still be interesting for a healthcare system?

Because early detection is almost always cheaper than late treatment. If I diagnose heart failure or an arrhythmia earlier, I can start treatment on an outpatient basis and avoid hospital stays. This saves costs and prevents worse outcomes.

This can also be seen in other examples, such as CT-supported heart screenings: AI analysis can detect deposits early and prevent heart attacks. Ultimately, this costs less than financing complex interventions later on.

In Germany, however, people seem to be more hesitant. Why is that?

One reason is the care structure. In Germany, we are very specialized: the family doctor refers you to a cardiologist, where ultrasound and ECG are available immediately. In other countries, these devices are not available everywhere—where an AI stethoscope can offer real added value.

On the other hand, the question arises as to how often a stethoscope is used in everyday life. Ultrasound is standard in everyday cardiological practice. GPs are also under time pressure. Listening to every patient in detail is hardly realistic. Such a stethoscope would be ideal for regions with less infrastructure: rural areas, developing countries, or places with long distances to a specialist. But the introduction is mainly taking place in countries with good care anyway. The additional benefit is less there.

Most of the AI stethoscopes currently available are also cloud-based. This means that the images are not evaluated locally but uploaded to servers where they are analyzed and then returned with a result. This poses several challenges. On the one hand, data protection. Medical data is particularly sensitive, and as soon as it is transferred, there must be absolute clarity about the storage location, encryption, and access rights. This complicates implementation, especially in a healthcare system with strict guidelines. Then there are the speed and costs. Every transfer requires bandwidth and server capacity. This is an additional expense in everyday practice and can be an obstacle, especially for GPs in rural areas or with poor infrastructure.

In the long term, it would make sense to increasingly rely on local or hybrid solutions in which some analysis can take place directly on the device itself. This would reduce dependence on cloud systems and make data protection issues easier to solve—and at the same time, the results could be returned more quickly.

As a cardiologist, what are you currently working on in the field of digitalization?

I am heavily involved in digital cardiology. For example, we are looking into research funding to see how apps can be used in everyday practice. Although the “apps on prescription” (DiGA) are available, they are difficult to implement in the field of cardiology. While improvements in quality of life can be quickly demonstrated for mental illnesses, in cardiology the only thing that often counts is fewer heart attacks and fewer deaths. This is difficult for a single app to prove.

We also see that people are happy to use health apps as long as they are healthy—as a kind of "gadget." But sick people have little desire to be reminded of their illness every day.

Cardiology is already very advanced in the field of telemedicine.

The first telemedicine projects—for example, the continuous care of pacemaker or heart failure patients—have already been transferred to standard care. Patients regularly measure their blood pressure or weight and automatically transmit their values to a telemedicine center. There, any abnormalities are passed on directly to the doctors treating them.

In rural areas in particular, telemedicine can improve care, avoid hospitalization, and make life easier for patients. AI stethoscopes could take on a similar role in the future—for example, by enabling preliminary screenings or directing patients with unclear symptoms to specialist care in a more targeted manner.

However, it is important that these systems do not stand alone but are closely linked to existing digital infrastructure (practice systems, electronic patient records, TI Messenger). This is the only way to ensure that the benefits for patients are great enough and that doctors are not burdened with additional work.

Two things are important here: firstly, we need sensible implementation in care and processes. Secondly, we need to clarify where the benefits are greatest—both medically and economically. AI alone will not solve health problems. But properly integrated, it could help to create fairness in care—especially where resources are lacking.

I currently see few areas of application for Germany because we have established more precise standard methods, such as ultrasound. Such devices could make more sense in regions without dense cardiological care.

(mack)