Smart stethoscope for cardiac auscultation in general practice: a prospective feasibility study of AI-assisted detection of atrial fibrillation, heart failure, and valvular heart disease

This prospective feasibility study demonstrates that AI-enabled digital stethoscopes are a practical and effective tool for detecting atrial fibrillation, heart failure, and valvular heart disease in routine general practice, achieving high rates of analyzable recordings with minimal workflow disruption.

The Gist

Imagine your heart is like a complex, high-performance engine. Usually, a doctor listens to this engine with a traditional stethoscope, which is like a basic mechanic's ear. They can hear if the engine is running smoothly or if there's a rattle, but it requires years of training to hear the subtle differences between a minor squeak and a serious breakdown.

This paper is about testing a new, "super-charged" mechanic's tool: a smart stethoscope powered by Artificial Intelligence (AI).

Here is the story of the study, broken down into simple concepts:

1. The Problem: The "Silent" Engine

Many heart problems (like irregular rhythms, weak pumps, or leaky valves) are like silent leaks in a car. They often don't show obvious symptoms until the car breaks down completely on the highway. By the time patients feel really sick, it's often too late to fix things easily. Doctors in regular clinics (General Practice) are busy and need a way to spot these "silent leaks" early, right in the waiting room.

2. The New Tool: The "Smart Detective"

The researchers tested a device called the Eko CORE 500. Think of this not just as a listening device, but as a detective with two superpowers:

• Super Hearing: It amplifies the heart sounds so much that even the quietest whispers of a heart valve can be heard.
• Super Vision (ECG): It also acts like a tiny camera that takes a picture of the heart's electrical rhythm at the same time.

The AI inside the device acts like a co-pilot. While the doctor listens, the AI instantly analyzes the sounds and the electrical signals, shouting out, "Hey, I think I hear an irregular rhythm!" or "Warning: Possible valve leak!"

3. The Experiment: A Real-World Test Drive

The researchers didn't test this in a quiet, perfect laboratory. Instead, they took the tool out for a real-world test drive in a busy Amsterdam clinic and even into people's homes.

• The Drivers: 50 older adults (aged 65+), the group most likely to have heart trouble.
• The Mission: The doctor used the smart stethoscope during normal check-ups. They tried to listen to the heart in four different spots, like checking all four corners of a car engine.

4. The Results: Did it Work?

The study asked two main questions: "Is it easy to use?" and "Is it accurate?"

• Ease of Use (Feasibility): Yes! The tool was surprisingly easy to fit into a normal doctor's visit. It only took about 1 to 2 minutes (the time it takes to brew a cup of coffee). It didn't slow down the doctor or confuse the patients.

  • The Hiccup: Just like a car radio can get static if you have too much hair on the antenna or if the skin is too oily, the device sometimes struggled with very obese patients, heavy chest hair, or large breasts. In about 6% of cases, the signal was too "noisy" to read. But in 94% of cases, the AI got a clear reading.

• Accuracy (Performance): The AI was a very good detective.

  • It correctly identified heart problems in 8 out of 10 patients who actually had them.
  • It rarely cried "wolf" when there was no danger (it was 92% specific).
  • The Bonus Catch: In one case, the AI heard a problem the doctor hadn't noticed yet. It flagged a leaky valve, and a follow-up scan confirmed the doctor needed to treat a patient who was feeling tired but didn't know why.

5. The Bottom Line

This study is like a proof-of-concept prototype. It shows that AI-powered stethoscopes aren't just sci-fi dreams; they can actually work in a busy, messy, real-world doctor's office.

The Metaphor for the Future:
Currently, the smart stethoscope is like a GPS navigation system for heart health. It doesn't drive the car for you (the doctor still makes the final decision), but it helps you see traffic jams (heart problems) you might have missed, ensuring you take the safest route to keep your patients healthy.

The researchers say we need bigger studies to be 100% sure, but this first test drive suggests that this technology could soon become a standard tool in every doctor's bag, helping catch heart trouble before it becomes a crisis.

Technical Summary

1. Problem Statement

Cardiovascular diseases (CVD), including atrial fibrillation (AF), heart failure (HF), and valvular heart disease (VHD), are leading causes of global morbidity and mortality. A critical gap exists in early detection:

• Late Diagnosis: These conditions are often diagnosed only after acute deterioration, limiting the efficacy of guideline-directed therapies.
• Primary Care Limitations: Patients frequently present with non-specific symptoms (e.g., dyspnea, fatigue) in general practice. Traditional cardiac auscultation relies heavily on clinician experience and is often insufficient in busy primary care settings to detect subtle structural or rhythm abnormalities.
• Need for Innovation: There is a need for point-of-care tools that enhance diagnostic accuracy without disrupting routine clinical workflows. While "smart stethoscopes" combining phonocardiography and electrocardiography (ECG) with AI exist, their real-world feasibility in community-based primary care and home visits remains under-explored.

2. Methodology

Study Design & Setting:

• Type: Prospective diagnostic feasibility study.
• Location: A large primary care center in Amsterdam, The Netherlands (including clinic consultations and home visits).
• Timeline: December 2025 – February 2026.
• Participants: 50 consecutive patients aged ≥65 years. No exclusion criteria were applied.
• Ethics: Waived formal review as part of a quality improvement initiative; patients had an opt-out option for de-identified data use.

Index Test (Intervention):

• Device: Eko CORE 500 digital stethoscope (FDA-cleared, CE-marked) paired with a smartphone application.
• Mechanism: Simultaneous recording of ECG and phonocardiography (heart sounds).
• Protocol: Recordings attempted at four standard positions:
  1. Right upper sternal border
  2. Left upper sternal border
  3. Left lower sternal border
  4. Apex
• AI Output: Automated classification of rhythm (specifically AF detection) and indication of suspected structural abnormalities.
• Optimization: If signal quality was poor, clinicians used alcohol wipes to remove skin lotions/creams and applied increased pressure for better contact.

Reference Standard:

• Diagnosis based on previously established records in the Electronic Medical Record (EMR) prior to inclusion (AF, HF, or VHD confirmed via routine ECG, imaging, or specialist correspondence).
• Note: This study did not use contemporaneous systematic testing (e.g., immediate echocardiography for all) as the reference standard, relying instead on existing clinical history.

Metrics:

• Feasibility: Technical failure rates, workflow disruption, acquisition time, and proportion of analyzable recordings.
• Diagnostic Performance: Sensitivity, specificity, and predictive values for detecting known AF, HF, or VHD.

3. Key Contributions

• Real-World Feasibility: This is one of the first studies to evaluate the integration of AI-assisted auscultation specifically within routine general practice and home visits, rather than controlled hospital environments.
• Workflow Integration: Demonstrated that the technology can be adopted without significant disruption to standard care, with acquisition times of 1–2 minutes.
• Signal Acquisition Challenges: Identified specific, practical barriers to signal quality in a primary care setting (obesity, chest hair, breast tissue, and skin lotions) and documented mitigation strategies (alcohol wipes, pressure adjustment).
• New Case Detection: Highlighted the potential for AI to identify previously undiagnosed, clinically significant pathology (a case of moderate-to-severe mitral regurgitation).

4. Results

Feasibility Outcomes:

• Completion: The procedure was completed in all 50 patients with no device malfunctions.
• Time: Median acquisition time was 1–2 minutes (max 5 minutes); no meaningful workflow disruption was reported.
• Analyzability:
  • At least one analyzable recording (successful AI output) was obtained in 47/50 patients (94%).
  • Complete four-position analysis was achieved in 42/50 patients (84%).
• Failure Causes: The 3 non-analyzable cases were due to obesity, chest hair, or excess breast tissue affecting acoustic coupling or ECG contact.

Diagnostic Performance (Exploratory):

• Cohort: Among the 47 analyzable patients, 11 had known diagnoses of AF, HF, or VHD.
• Accuracy:
  • Sensitivity: 81.8% (9/11 true positives detected).
  • Specificity: 91.7% (High negative predictive value).
• Error Analysis:
  • False Negatives (2): One case of paroxysmal AF (patient not in AF during recording) and one case of HF with preserved ejection fraction.
  • False Positives (3): One clinically insignificant murmur; two AF alerts triggered by rhythm irregularity and baseline noise.
• Clinical Impact: The AI system prompted further evaluation in one patient presenting with fatigue, leading to an echocardiogram that confirmed moderate-to-severe mitral regurgitation (a new diagnosis).

5. Significance and Conclusion

• Operational Viability: The study confirms that AI-enabled digital stethoscopes are operationally feasible in routine primary care and home settings for older adults, with high rates of successful data capture.
• Clinical Utility: The high specificity and negative predictive value suggest the tool is effective for "ruling out" conditions, while the identification of a new case of valvular disease demonstrates its potential for early detection.
• Limitations: The study is limited by its small sample size, single-site design, and the use of historical EMR diagnoses rather than contemporaneous gold-standard testing. The GP was not blinded to patient history, which may introduce bias.
• Future Directions: The authors conclude that while the technology is promising, larger, multi-center pragmatic trials with systematic reference standards are required to definitively establish clinical effectiveness and impact on patient outcomes.

Summary Verdict: The paper provides strong preliminary evidence that AI-assisted cardiac auscultation can be seamlessly integrated into general practice, offering a feasible, rapid, and accurate method to screen for major cardiac conditions in older populations, provided that signal acquisition challenges related to patient anatomy are managed.