SonicGuard Sensor - A Multichannel Acoustic Sensor for Long-Term Monitoring of Abdominal Sounds Examined through a Qualification Study

Zahra Mansour, Verena N. Uslar, Dirk Weyhe, Danilo Hollosi, Nils Strodthoff

TL;DR

This paper introduces SonicGuard, a novel, lightweight, multichannel acoustic sensor designed for long-term monitoring of abdominal sounds, particularly bowel sounds. Through a rigorous qualification study involving phantoms and real patients, SonicGuard demonstrates performance comparable to gold-standard digital stethoscopes, paving the way for advanced machine learning analysis of bodily sounds.

What Problem Does This Paper Address?

Auscultation, the practice of listening to internal body sounds, is a cornerstone of medical diagnosis. However, traditional stethoscopes have limitations: they are often single-channel, require significant user expertise for accurate interpretation, and are not well-suited for continuous, long-term monitoring. While digital stethoscopes have emerged, they can be bulky, expensive, and still often lack the multi-channel capability needed to capture complex spatial information from bodily sounds, such as the distinct sounds originating from different abdominal quadrants. This gap is particularly critical for monitoring bowel sounds, which are intermittent and require prolonged observation for accurate assessment. The authors aim to address these limitations by developing a more accessible, versatile, and data-rich acoustic sensing solution.

Methods

The core of this work is the development and evaluation of the SonicGuard sensor. This is a multichannel acoustic sensor designed for long-term recording of bodily sounds. Key features include:

  • Lightweight and Compact Design: Significantly smaller and lighter than commercial digital stethoscopes, enhancing comfort for extended monitoring.
  • Multichannel Capability: Allows for simultaneous recording from multiple locations, crucial for spatial analysis of sounds.
  • Low Self-Noise: Ensures high-quality audio capture even in quiet environments.
  • IP57 Water and Dust Resistance: Enhances durability for various environmental conditions.
  • Affordable Components: Utilizes a 3D-printed housing and standard components to reduce production costs, enabling the deployment of multiple sensors.

The qualification study involved a multi-stage approach:

  1. Basic Qualification Tests: Individual SonicGuard sensors were tested against a calibrated reference microphone in an isolated room to assess reliability, delay, RMS error, and frequency response.
  2. Gelatin Box Phantom: A custom-built gelatin box phantom was used to simulate the acoustic properties of abdominal layers. This allowed for controlled testing of frequency response, self-noise, and resilience to air gaps (simulating imperfect contact).
  3. iSTAN Healthcare Phantom: A sophisticated medical phantom was employed to simulate various bowel, heart, and lung sounds. This allowed for testing of sensor performance across different sound types, quadrants, and the evaluation of machine learning capabilities for classifying bowel sound states (normal, hyperactive, hypoactive).
  4. Real Patient Tests: Healthy subjects were used to assess sensor performance during simultaneous recording and to evaluate user experience and comfort.

Datasets

The study utilized several datasets for evaluation:

  • Simulated Datasets:
    • Gelatin Box Phantom Data: Generated by playing a reference bowel sound signal through the phantom.
  • iSTAN Phantom Data: Generated by simulating normal, hyperactive, and hypoactive bowel sounds, as well as normal heart and lung sounds. This data was used for both descriptive analysis and machine learning model training.
  • Real Patient Data:
    • Abdominal Sound Recordings: Collected from 10 healthy subjects for simultaneous sensor comparison.
  • Usability Survey Data: Collected from another 10 subjects to assess user experience and comfort.

The iSTAN phantom data was specifically structured for machine learning, creating datasets of 1-minute recordings for each of the three bowel sound states, split into 6-second segments.

Key Insights and Ideas

  • The Power of Multichannel Sensing: The ability to record from multiple locations simultaneously is a significant advantage for understanding the spatial distribution and origin of bodily sounds, which is difficult with traditional single-channel stethoscopes.
  • Machine Learning Readiness: The SonicGuard sensor's high-quality data output, coupled with its affordability, makes it an ideal candidate for building large datasets for training advanced machine learning models to diagnose conditions based on bodily sounds.
  • Comfort and Usability for Long-Term Monitoring: The sensor's small size and light weight are crucial for enabling comfortable, long-term monitoring, which is essential for capturing intermittent events like bowel sounds.
  • Phantom Studies as a Bridge: The use of both simple gelatin phantoms and more complex iSTAN phantoms provides a valuable intermediate step between controlled lab environments and real-world patient data, allowing for systematic evaluation and validation.

Key Results

  • Comparable Performance to Gold Standard: The SonicGuard sensor demonstrated acoustic performance comparable to the Littmann CORE and Thinklabs One digital stethoscopes across various tests, including frequency response and self-noise.
  • High Reliability: Basic qualification tests showed negligible differences between six individual SonicGuard sensors, indicating high reliability.
  • Resilience to Air Gaps: SonicGuard, along with Littmann, showed greater resilience to a 2mm air gap compared to Thinklabs, maintaining performance better under imperfect contact conditions.
  • Machine Learning Accuracy: In machine learning tasks to classify bowel sound states (normal, hyperactive, hypoactive), the SonicGuard sensor consistently achieved the highest classification accuracy, reaching up to 92.7% with an SVM classifier.
  • Usability and Comfort: In patient surveys, 40% of participants favored the SonicGuard for comfort, and 63% preferred it for extended recording sessions, highlighting its suitability for long-term monitoring.

Limitations and Future Work

  • Limited Scope of Qualification: While the study focused on abdominal sounds, the authors acknowledge that a full qualification study for heart and lung sounds is beyond the scope of this work, though preliminary tests showed promising results.
  • Phantom vs. Real-World Complexity: While phantoms are valuable for controlled testing, they may not perfectly replicate the full complexity and variability of sounds encountered in real patients.
  • Future Work: The authors envision integrating the SonicGuard sensor with user-friendly mobile hardware and advanced machine learning modules to create a comprehensive system for effectively detecting anomalies in bowel sounds and other bodily sounds. This could lead to improved early disease detection and patient monitoring, particularly after surgical procedures. Further exploration of its application for lung and heart sounds is also planned.

Summary updated

Abstract

Auscultation is a fundamental diagnostic technique that provides valuable diagnostic information about different parts of the body. With the increasing prevalence of digital stethoscopes and telehealth applications, there is a growing trend towards digitizing the capture of bodily sounds, thereby enabling subsequent analysis using machine learning algorithms. This study introduces the SonicGuard sensor, which is a multichannel acoustic sensor designed for long-term recordings of bodily sounds. We conducted a series of qualification tests, with a specific focus on bowel sounds ranging from controlled experimental environments to phantom measurements and real patient recordings. These tests demonstrate the effectiveness of the proposed sensor setup. The results show that the SonicGuard sensor is comparable to commercially available digital stethoscopes, which are considered the gold standard in the field. This development opens up possibilities for collecting and analyzing bodily sound datasets using machine learning techniques in the future.
BibTeX
@inproceedings{mansour2024sonicguard,
  author = {Zahra Mansour and Verena N. Uslar and Dirk Weyhe and Danilo Hollosi and Nils Strodthoff},
  title = {{SonicGuard Sensor - A Multichannel Acoustic Sensor for Long-Term Monitoring of Abdominal Sounds Examined through a Qualification Study}},
  booktitle = {Sensors},
  year = {2024},
  url = {https://openreview.net/forum?id=PbZFFTJ1yM},
}
Figure