Simulation-based optimisation to quantify heterogeneity of specific ventilation and perfusion in the lung by the Inspired Sinewave Test

Minh C. Tran, Vu Nguyen, Richard Bruce, Douglas C. Crockett, Federico Formenti, Phi Anh Phan, Stephen J. Payne, Andrew D. FarmeryScientific 2021 · DOI

Abstract

The degree of specific ventilatory heterogeneity (spatial unevenness of ventilation) in the lung is a critical marker for early structural changes and disease detection. The Inspired Sinewave Test (IST) is a non-invasive respiratory test capable of measuring cardiopulmonary parameters. This study presents a simulation-based optimization framework for the IST, utilizing a realistic lung model with lognormal distributions of spatial ventilation and perfusion. Validated in both animal models (13 pigs) and human subjects (104 participants), the method successfully quantified heterogeneity, allowing for diagnostic distinction between healthy and diseased states. Key findings include a linear relationship between ventilatory heterogeneity and age in healthy individuals, high sensitivity to lung injury in animal models ($p < 0.0001$), and effective identification of COPD patients (AUC 0.76) with 73% agreement with spirometry for phenotyping.

Introduction: The Challenge of Early Detection

In a healthy lung, air (ventilation) and blood (perfusion) are distributed relatively evenly across the alveolar compartments. However, conditions such as Chronic Obstructive Pulmonary Disease (COPD) and Acute Respiratory Distress Syndrome (ARDS) cause structural changes that result in a "patchy" or heterogeneous distribution. Detecting these changes early is difficult, as they often precede abnormalities in standard tests like spirometry by decades. While advanced imaging (CT/MRI) can visualize this heterogeneity, it is often too costly or invasive for routine bedside use. The Inspired Sinewave Test (IST) offers a simpler, non-invasive alternative by analyzing how the lung processes a tracer gas.

Methods: Simulation-Based Optimization

The core innovation of this research is a simulation-based optimization framework that translates IST data into quantitative maps of lung heterogeneity.

Optimization workflow and lung heterogeneity distributions for healthy, emphysema, and pulmonary embolism states.

As illustrated in Figure 1a, the process begins with an initial parameter set and iteratively runs a lung simulation. During the IST, patients breathe a tiny, oscillating amount of nitrous oxide ($N_2O$). The device measures the lung's filtering effect on this sine wave. The researchers modeled the lung as a collection of 125 alveolar compartments, assuming that the distribution of ventilation and perfusion follows a lognormal distribution.

To handle the complexity of this non-convex model, the team employed Bayesian Optimization (BO). This approach efficiently searches for the optimal distribution parameters—$\sigma_V$ for ventilation and $\sigma_P$ for perfusion—that best match the patient's real-world data. This method significantly reduced processing time to under 15 minutes, making it viable for clinical environments.

Key Insights and Results

The framework was tested on 13 anesthetized pigs (pre- and post-lung injury) and 104 human subjects, including healthy volunteers and COPD patients across various GOLD stages.

The Significance of Distribution "Width" The study revealed that the mean values of air and blood flow were less indicative of disease than the standard deviation ($\sigma$) of their distributions. As shown in Figures 1b-d, healthy lungs exhibit narrow, peaked distributions ($\sigma_V=0.3$), while diseased states like emphysema ($\sigma_V=1.2$) or pulmonary embolism ($\sigma_P=1.3$) show much wider, more heterogeneous curves.

Diagnostic Performance * Age-Related Changes: In healthy subjects, ventilatory heterogeneity increased linearly with age ($R^2 = 0.42$), reflecting natural lung aging. * COPD Phenotyping: The IST-based indices identified COPD with an AUC of 0.76 and matched standard spirometry phenotypes with 73% agreement. * Sensitivity to Injury: In animal models, the test showed highly significant differences ($p < 0.0001$) between healthy and injured lungs, regardless of variations in positive end-expiratory pressure (PEEP). * Computational Success: Bayesian Optimization achieved a success rate of 95% in animal data and 82% in human data, outperforming traditional optimization methods.

Limitations and Future Work

Despite its success, the test is naturally more sensitive to ventilation than perfusion due to the modest blood solubility of $N_2O$. Consequently, perfusion results ($\sigma_P$) are generally less reliable. Furthermore, the current model assumes a single lognormal distribution, which may not capture the multi-modal distributions seen in some complex pathologies. Future research will focus on validating these IST-derived indices against high-resolution imaging like CT scans to confirm the spatial accuracy of the identified heterogeneity.

Summary updated

Abstract

The degree of specific ventilatory heterogeneity (spatial unevenness of ventilation) of the lung is a useful marker of early structural lung changes which has the potential to detect early-onset disease. The Inspired Sinewave Test (IST) is an established noninvasive 'gas-distribution' type of respiratory test capable of measuring the cardiopulmonary parameters. We developed a simulation-based optimisation for the IST, with a simulation of a realistic heterogeneous lung, namely a lognormal distribution of spatial ventilation and perfusion. We tested this method in datasets from 13 anaesthetised pigs (pre and post-lung injury) and 104 human subjects (32 healthy and 72 COPD subjects). The 72 COPD subjects were classified into four COPD phenotypes based on 'GOLD' classification. This method allowed IST to identify and quantify heterogeneity of both ventilation and perfusion, permitting diagnostic distinction between health and disease states. In healthy volunteers, we show a linear relationship between the ventilatory heterogeneity versus age ([Formula: see text]). In a mechanically ventilated pig, IST ventilatory heterogeneity in noninjured and injured lungs was significantly different (p < 0.0001). Additionally, measured indices could accurately identify patients with COPD (area under the receiver operating characteristic curve is 0.76, p < 0.0001). The IST also could distinguish different phenotypes of COPD with 73% agreement with spirometry.
BibTeX
@inproceedings{tran2021simulation,
  author = {Minh C. Tran and Vu Nguyen and Richard Bruce and Douglas C. Crockett and Federico Formenti and Phi Anh Phan and Stephen J. Payne and Andrew D. Farmery},
  title = {{Simulation-based optimisation to quantify heterogeneity of specific ventilation and perfusion in the lung by the Inspired Sinewave Test}},
  booktitle = {Scientific},
  year = {2021},
  doi = {10.1038/s41598-021-92062-w},
}
Figure