⚡ electrical engineering

Comprehensive Dataset and Signal Processing Framework for Phonocardiogram-Based Heart Rate and Blood Pressure Estimation

This paper introduces a cost-effective Phonocardiogram Tracking (PhonoTrack) system and a corresponding dataset that enable non-invasive estimation of both heart rate and blood pressure using only PCG signals, achieving high accuracy through advanced signal processing and semi-empirical modeling.

Original authors: Abdul Ahad Mamun, Utsab Saha, Md Hasibul Hasan, Shahed Ahmed, MD Jahin Alam

Published 2026-05-25

📖 4 min read☕ Coffee break read

CC BY 4.0

Original authors: Abdul Ahad Mamun, Utsab Saha, Md Hasibul Hasan, Shahed Ahmed, MD Jahin Alam

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine your heart as a busy, rhythmic drumming machine inside your chest. Every time it beats, it makes a specific "thump-thump" sound (the heart sounds) and sends a tiny electrical spark (the electrical signal). Usually, to check how fast your heart is beating or what your blood pressure is, doctors need expensive machines, sticky wires, or a tight cuff that squeezes your arm.

This paper introduces a new, low-cost invention called PhonoTrack. Think of it as a "smart stethoscope" that tries to figure out your heart rate and blood pressure just by listening to the drumming sounds, without needing any electrical wires or arm cuffs.

Here is how they did it, explained simply:

1. The Setup: Building the "Smart Ears"

The researchers built a simple device using a standard stethoscope tube with a microphone hidden inside the earpiece. They strapped this to the chest of 15 healthy volunteers (all men, aged 18–50).

The Goal: To record the heart's "thump-thump" sounds (called PCG) and compare them to a "gold standard" reference.
The Reference: At the same time, they recorded the heart's electrical signals (ECG) using standard wires and measured blood pressure manually with a traditional arm cuff. This was their "answer key" to see if their new device was correct.

2. Cleaning the Signal: The "Noise Canceling" Headphones

Heart sounds are often messy, like trying to hear a whisper in a windy room. The computer had to clean up the recording.

They used a digital "sieve" (called Wavelet Denoising) to filter out the wind and static noise, leaving only the clear heartbeats.
They then used three different "mathematical lenses" to find the peaks of the heartbeats:
1. Hilbert Transform: Like tracing the outline of a wave to see its height.
2. Shannon Entropy: Like turning up the volume on the loudest parts of the sound to make the beats stand out.
3. Wavelet Energy Spectrum: Like looking at a heat map of the sound to see where the energy is concentrated.

3. Counting the Beats (Heart Rate)

Once the sounds were clean, the computer counted the "thumps."

The Result: The new device was incredibly accurate. It matched the "gold standard" electrical wires almost perfectly.
The Score: If the perfect score is 1.0, the device scored 0.973 (using the Shannon Entropy method). This is like a student getting an A+ on a test. The difference between the new device and the gold standard was less than 2 beats per minute on average.

4. Guessing the Pressure (Blood Pressure)

This is the tricky part. Usually, you need a cuff to squeeze the arm to measure pressure. This device tried to guess the pressure just by listening to the shape and timing of the heart sounds.

The Analogy: Imagine a drummer. If they hit the drum harder and faster, the sound might change slightly. The researchers built a "mathematical translator" (a semi-empirical model) that looked at how long the "thump" lasted, how fast it rose, and how fast it fell. They fed these details into a formula to guess the blood pressure.
The Result:
- Systolic (Top number): The guess was very close to the real measurement, with a correlation of 0.89.
- Diastolic (Bottom number): The guess was decent but a bit more variable, with a correlation of 0.70.
- Note: The paper admits this part is a "proof of concept." It works well in a quiet lab, but it's not perfect yet.

5. The Catch (Limitations)

The paper is very honest about what it doesn't do yet:

Small Group: They only tested 15 men. They didn't test women, children, or older people.
Quiet Room: The tests were done in a silent lab. They haven't tested it while someone is walking, running, or in a noisy street.
Not a Doctor Yet: The authors explicitly state this is a prototype. It is not ready to replace a doctor's equipment in a hospital or for diagnosing heart disease. It is a "first step" to see if a cheap, portable device could work.

Summary

The researchers built a cheap, portable device that listens to your heart. They proved that by using clever math to clean up the sound, they can count your heartbeats almost as accurately as expensive medical wires. They also showed a promising (but not perfect) way to guess your blood pressure just by listening. It's a successful "lab experiment" that suggests a future where you might be able to check your heart health with a simple, low-cost gadget, but more testing with more people is needed before it can be used in the real world.

Technical Summary: Comprehensive Dataset and Signal Processing Framework for Phonocardiogram-Based Heart Rate and Blood Pressure Estimation

Problem Statement

Cardiovascular diseases (CVDs) are a leading cause of global mortality, often termed "silent killers" due to their difficulty in early detection. While advanced imaging (CT, CMRI) and multimodal monitoring (ECG combined with PPG) exist, they face limitations regarding cost, operational complexity, and portability. Specifically, ECG-PPG systems for estimating Heart Rate (HR) and Blood Pressure (BP) via Pulse Arrival Time (PAT) are susceptible to lead misalignment and insufficient for detecting specific heart abnormalities like murmurs. Furthermore, existing Phonocardiogram (PCG) studies often rely on complex machine learning models unsuitable for portable devices or fail to simultaneously estimate both HR and BP using PCG signals alone. There is a critical need for a low-cost, non-invasive, and portable system capable of continuous cardiovascular monitoring using only acoustic heart sounds.

Methodology

1. Data Acquisition and Dataset Construction

The authors developed a custom, low-cost "PhonoTrack" system to capture synchronized PCG and Electrocardiogram (ECG) signals alongside manual blood pressure readings.

Hardware: The PCG signal was captured using a microphone embedded in a standard stethoscope diaphragm, connected to an iPad for recording (44.1 kHz). The ECG was captured via a three-lead AD8232 sensor module connected to an Arduino.
Protocol: Data was collected from 15 healthy male volunteers (aged 18–50) in a controlled indoor environment. Each session lasted 60 seconds.
Ground Truth: HR was validated against the ECG signal (R-peaks), and BP was validated against manual sphygmomanometer readings taken immediately before and after the recording session.
Dataset: The resulting dataset includes 15 synchronized PCG-ECG recordings with paired cuff BP values, providing time-aligned data for benchmarking.

2. Signal Processing Framework

The study employs a unified framework of standard Digital Signal Processing (DSP) techniques rather than complex deep learning models.

Heart Rate (HR) Estimation:

Preprocessing: PCG signals underwent wavelet-based denoising (Daubechies db4, 8 levels) and normalization.
Envelope Extraction: Three distinct methods were used to extract the signal envelope to identify S1 and S2 heart sounds:
1. Hilbert Transform (HT): Extracts instantaneous amplitude.
2. Shannon Entropy (SE): Emphasizes high-energy regions to sharpen transient components.
3. Wavelet Energy Spectrum (WES): Analyzes energy distribution across frequency bands.
Peak Detection: Local maxima in the smoothed envelopes (filtered with a 20 Hz low-pass Butterworth filter) were identified as S1 and S2 peaks. Constraints were applied (min peak distance 0.125s, min height 15%) to avoid false detections.
Calculation: HR was derived from the average S1–S1 (or S2–S2) intervals.

Blood Pressure (BP) Estimation:

Feature Extraction: A set of physiologically interpretable timing features was extracted from the PCG envelope, including:
- Systolic ( $t_{sys}$ ) and Diastolic ( $t_{dias}$ ) durations.
- Rise and decay times for S1 and S2.
- Peak durations of S1 and S2.
- Estimated HR ( $HR_{pcg}$ ).
Modeling: A semi-empirical multiple regression model was developed to map these features to Systolic (SBP) and Diastolic (DBP) blood pressure. The model includes linear, quadratic, and interaction terms.
Validation: The model was evaluated using subject-wise Leave-One-Subject-Out Cross-Validation (LOOCV) to assess generalizability and prevent overfitting given the small sample size.

Key Contributions

PhonoTrack System: Development of a cost-effective, portable hardware prototype capable of simultaneous PCG and ECG acquisition.
New Dataset: Introduction of a synchronized PCG-ECG dataset with paired manual BP measurements from 15 subjects, addressing the lack of time-aligned PCG-BP ground truth in existing literature.
Algorithmic Framework: Demonstration that standard DSP techniques (HT, SE, WES) combined with semi-empirical regression can effectively estimate both HR and BP from PCG signals alone, avoiding the need for complex multimodal sensors or heavy machine learning models.
Comprehensive Validation: Rigorous evaluation including Pearson correlation, Root Mean Square Error (RMSE), Bland-Altman analysis, and Signal-to-Noise Ratio (SNR) assessment.

Results

Heart Rate Estimation:

All three envelope detection methods showed strong agreement with ECG-derived HR.
Shannon Entropy (SE) performed best, achieving a Pearson correlation coefficient ( $R$ ) of 0.973, with a Mean Absolute Error (MAE) of 1.330 bpm and RMSE of 1.688 bpm.
Hilbert Transform (HT) and Wavelet Energy Spectrum (WES) achieved $R$ values of 0.965 and 0.955, respectively.
Bland-Altman analysis confirmed low systematic bias and tight agreement within 95% limits of agreement.

Blood Pressure Estimation:

The semi-empirical regression model demonstrated moderate to strong correlation with ground truth BP.
Systolic BP (SBP): $R = 0.891$ , with a standard deviation of error of 2.10 mmHg.
Diastolic BP (DBP): $R = 0.700$ , with a standard deviation of error of 3.20 mmHg.
Under subject-wise LOOCV, the model yielded an SBP RMSE of 9.89 mmHg and DBP RMSE of 11.05 mmHg, indicating the model's current limitations in generalizing to unseen subjects with the current small cohort.

Significance and Claims

The paper positions the PhonoTrack system as a promising proof-of-concept for non-invasive, continuous cardiovascular monitoring. The authors claim that their approach offers a low-cost, portable, and single-sensor alternative to complex multimodal systems (ECG+PPG) or expensive imaging techniques.

However, the authors maintain a modest and cautious stance regarding clinical readiness:

They explicitly state that the findings are preliminary and based on a small, homogeneous cohort (15 healthy males).
They acknowledge that the current results are insufficient to support claims related to clinical deployment, routine screening, or cardiovascular disease assessment.
The study emphasizes that the semi-empirical model serves as a baseline, and future work must involve larger, more diverse populations (including females and various age/BMI groups) and real-world conditions (motion, noise) to establish broader generalizability.

In summary, the paper validates the feasibility of using PCG-only signals for HR and BP estimation using standard signal processing, laying the groundwork for future development of affordable wearable cardiac monitors, while clearly delineating the gap between current experimental results and clinical application.