An Apparatus for the Simulation of Breathing Disorders: Physically Meaningful Generation of Surrogate Data

Harry J. Davies, Ghena Hammour, Danilo P. Mandic, Mandic, Danilo P.arXiv 2021 · DOI · arXiv

Introduction

The rising prevalence of chronic respiratory conditions like Chronic Obstructive Pulmonary Disease (COPD) has spurred interest in using Artificial Intelligence (AI) for early detection and monitoring. However, the development of robust AI models is often hindered by the scarcity of high-quality, diverse patient data. To bridge this gap, researchers at Imperial College London have developed a low-cost, physical apparatus designed to simulate various breathing disorders in healthy subjects, generating "physically meaningful" surrogate data for machine learning.

The Challenge of Data Scarcity in Respiratory AI

Deep learning models are notoriously data-hungry. Collecting respiratory waveforms from a broad spectrum of patients is logistically complex and expensive. While synthetic data generation (e.g., via GANs) is an option, these models often lack "domain awareness"—they may mimic the look of a waveform without adhering to the underlying physiological mechanics. There is a critical need for surrogate data that captures the actual physics of respiratory obstruction and restriction.

Hardware Design and Methodology

The proposed system uses a physical "breathing circuit" that healthy subjects breathe through to simulate disease states.

Schematic and physical realization of the breathing disorder simulation apparatus, highlighting the mouth input, one-way valves, variable diameter tubes, and flow meter.

The apparatus consists of several key components:

  • One-Way Valves: 3D-printed cone funnels with foam plugs act as check valves, effectively separating the inspiratory and expiratory paths.
  • Variable Resistance Tubes: By swapping PVC tubes of different diameters (from 25mm for "healthy" down to 3mm for "severe obstruction"), researchers can independently control the resistance during each phase of the breath.
  • Data Acquisition: A digital flow meter (Sensirion SFM3200) connected to an Arduino Uno samples airflow at 20Hz.
  • Cross-Modal Sensing: Photoplethysmography (PPG) sensors are used to record blood flow changes modulated by thoracic pressure, providing a secondary data stream.

Key Findings and AI Validation

The apparatus successfully simulated the clinical spectrum of COPD, producing FEV1/FVC ratios (the gold standard for diagnosis) ranging from healthy (0.98) to severe (0.12).

Respiratory waveforms and statistical metrics generated by the apparatus, showing changes in flow rate, duty cycle, amplitude ratio, and sample entropy across different resistance levels.

Several key insights emerged from the data:

  • Waveform Fidelity: For severe restriction (8mm:3mm ratio), the median inspiratory duty cycle was 34.4%, closely matching the ~35% duty cycle observed in clinical COPD patients.
  • Predictability as a Biomarker: As obstruction increased, the "Sample Entropy" of the signal decreased, meaning the breathing became more rhythmic and predictable—a known clinical indicator of respiratory distress.
  • AI Transferability: Most significantly, a Convolutional Neural Network (CNN) trained exclusively on this surrogate data was able to accurately classify real-world COPD patients. This demonstrates that the hardware-generated data is physically meaningful enough to generalize to clinical settings.

Conclusion and Future Directions

This apparatus provides a simple, effective, and physically grounded method for generating surrogate respiratory data. While current limitations include system leakage at high pressures and the use of constant rather than dynamic resistance, the results prove that "human-in-the-loop" physical simulation is a viable path forward for training AI in healthcare. Future work will focus on refining the hardware to capture dynamic airway changes and expanding the use of PPG for wearable-based home monitoring.

Summary updated

Abstract

The rapidly increasing prevalence of debilitating breathing disorders, such as chronic obstructive pulmonary disease (COPD), calls for a meaningful integration of artificial intelligence (AI) into healthcare. While this promises improved detection and monitoring of breathing disorders, AI techniques are almost invariably "data hungry" which highlights the importance of generating physically meaningful surrogate data. Indeed, domain aware surrogates would enable both an improved understanding of respiratory waveform changes with different breathing disorders, and enhance the training of machine learning algorithms. To this end, we introduce an apparatus comprising of PVC tubes and 3D printed parts as a simple yet effective method of simulating both obstructive and restrictive respiratory waveforms in healthy subjects. Independent control over both inspiratory and expiratory resistances allows for the simulation of obstructive breathing disorders through the whole spectrum of FEV1/FVC spirometry ratios (used to classify COPD), ranging from healthy values to values seen in severe chronic obstructive pulmonary disease. Moreover, waveform characteristics of breathing disorders, such as a change in inspiratory duty cycle or peak flow are also observed in the waveforms resulting from use of the artificial breathing disorder simulation apparatus. Overall, the proposed apparatus provides us with a simple, effective and physically meaningful way to generate faithful surrogate breathing disorder waveforms, a prerequisite for the use of artificial intelligence in respiratory health.
BibTeX
@misc{davies2021apparatus,
  author = {Harry J. Davies and Ghena Hammour and Danilo P. Mandic and Mandic, Danilo P.},
  title = {{An Apparatus for the Simulation of Breathing Disorders: Physically Meaningful Generation of Surrogate Data}},
  year = {2021},
  eprint = {2109.06699},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2109.06699},
}
Figure