Abstract
Pulmonary conditions rank among the top causes of illness and disability worldwide. Currently, the standard approach to assess a patient's condition involves visiting hospitals or specialist clinics for a clinical assessment. However, access to medical services is limited for under-resourced areas and during pandemics. We propose that pulmonary disease might influence vocalization, potentially aiding in the characterization of the condition. Computational fluid dynamics (CFD) presents significant potential for evaluating flow-induced acoustics within the lungs. In this dissertation, a potential novel method is explored to assess pulmonary disease by analyzing changes in a patient's cough. Specifically, the study investigates the feasibility of using computational fluid dynamics to examine flow-induced vocalizations, cough, and lung-generated acoustics for diagnosing pulmonary conditions. Studies are conducted under both healthy and diseased conditions; diseased conditions include three groups of patients whose lungs have: pneumonic changes; bronchiectacic lungs; and lungs with cavitary tuberculosis. The hypothesis is that a computational fluid dynamics model for evaluating flow-induced acoustics will accurately simulate the acoustics in both healthy and diseased lungs. Potential differences in fluid and acoustic behaviour between these pathologies will be modelled, tested, and described. Twenty-two vocal recording samples from infected patients admitted to hospital with clinical diagnoses that suggested pneumonia, bronchiectasis and cavitary tuberculosis (without apparent clinical overlap), were obtained and analyzed along a frequency versus sound pressure level chart. Computational fluid dynamics is employed to simulate flow distribution and capture the acoustics associated with various pathologies and scenarios. A lung geometry mesh was provided by North Carolina State University. Two methods were used to account for the flow-induced acoustics and turbulence, firstly, the Large Eddy Simulation (LES) with the Ffowcs Williams and Hawkings Model and the Realizable k-𝜀 with the Broadband Noise Source Model. Findings were compared with the vocal recordings and suggest that computational fluid dynamics can differentiate vocalizations between healthy lungs and those affected by pneumonia, bronchiectasis, and cavitary tuberculosis. Differences in sound pressure levels are observed across the frequency range from 0 kHz to 16 kHz. The findings from the cough recordings indicated that the sound pressure levels were highest for healthy lungs, followed by lungs with cavitary tuberculosis, lungs with pneumonic changes and bronchiectacic lungs, respectively. Similarly, our modelled data indicates that the sound pressure level (SPL) in decibels (dB) for cavitary tuberculosis and healthy lungs is higher across the frequency range when compared to bronchiectasis and pneumonia. Specifically, the healthy lungs maintain a higher sound pressure level across most of the frequency range compared to the pathological conditions and the curve steadily decreases with increasing frequency but has less pronounced dips, showing more stability in high frequencies. Cavitary tuberculosis follows a relatively smoother path but has a more rapid decline starting from around 6 kHz. Bronchiectasis shows more significant oscillations between 0 – 4 kHz with the response declining
4
rapidly and flattening out beyond 8 kHz indicating poor representation of higher frequencies. Pneumonia has more dips especially in the mid frequency 3 – 7 kHz which indicates significant sound attenuation. All the pathological conditions exhibit a more rapid sound pressure level drop-off compared to healthy lungs, especially in the higher frequency range. Cough recordings were compared to the computational fluid dynamics findings that we modelled with the Large Eddy Simulation (LES) with the Ffowcs Williams and Hawkings Model and the Realizable k-𝜀 with the Broadband Noise Source Model methods. To evaluate the accuracy of the model against the recorded data, statistical hypothesis testing was conducted using the reduced chi-squared method. While the turbulence and acoustics models did not produce identical results to the cough recording data, they exhibited similar qualitative correlations. The Large Eddy Simulation combined with the Ffowcs Williams and Hawkings Model for healthy lungs, pneumonic-infected lungs and lungs with bronchiectasis demonstrated qualitative correlation with the cough recordings. Similarly, the Realizable k-𝜀 with the Broadband Noise Source Model for healthy lungs, lungs with pneumonic changes and lungs with bronchiectasis also showed qualitative correlation to the cough recordings. The findings indicate that computational fluid dynamics could improve the understanding of flow-induced acoustics associated with various lung pathologies. Such simulations could be conducted under various conditions with the aim of creating synthetic datasets that can be used to train an artificial intelligence (AI) model/system with the objective of using artificial intelligence infused telemedicine-based diagnosis.