Determination of Respiratory Parameters during CPET Using Acoustic Signal Analysis
Qi Zhang (1,3), Christina De Oliviera Francisco (2,3), Muammar Kabir (2,3), Nasim Montazeri (2,3), Azadeh Yadollahi (2,3)
Department of Electrical & Computer Engineering, University of Toronto
Kite-Toronto Rehabilitation Institute, University Health Network
Institute of Biomaterial & Biomedical Engineering, University of Toronto
Introduction: Cardiovascular and respiratory diseases are leading causes of death. In this regard, cardiopulmonary exercise test (CPET) is the gold standard diagnostic tool. However, the application of CPET is expensive resulting in high rate of undiagnosed patients. On the other hand, biosignal analysis has become a convenient modality for clinical diagnosis. Ambient noises removal remains a challenge for clinical parameters determination. In this study, we aim to extract respiratory signal from tracheal sounds recorded during CPET and estimate respiratory parameters.
Methods: Thirty healthy subjects performed stepwise incremental exercise protocol on a stationary cycle ergometer. Participants’ respiratory function was continuously monitored by the gas exchange analysis system. The measurements were used as the reference. Simultaneously, we recorded participants’ breathing sounds over the suprasternal notch. The sound recordings were de-noised using a Synchrosqueezing-transform-based algorithm. Onsets of breath phases were sequentially detected using the energy of respiratory signal, calculated as the logarithm of its variance. Consequently, respiratory rate was calculated as the number of breathes per minute. The power intensity of each breath was calculated as the square of signal amplitude. We applied Bland-Altman analysis to validate the accuracy of respiratory rate determination. Moreover, correlations between the respiratory power intensity and ventilatory parameters were analyzed using pairwise Pearson’s or Spearman correlation test.
Results: One recording was excluded due to low signal quality. Analysis of 29 participants (8 males, age: 31.17±8.13 years, BMI: 24.09±3.53 kg/m2) showed that errors between the respiratory rate determined from the algorithm and the CPET system were within 1.2 breaths per minute. In addition, high correlations were found between the power intensity of respiratory signal and the ventilatory parameters: oxygen uptake (r > 0.75), carbon dioxide production (r > 0.80), tidal volume (r > 0.80), and minute ventilation (r > 0.75).
Conclusion: The proposed algorithm extracted respiratory signal from the contaminated recordings and accurately estimated respiratory rate. In addition, power intensity of respiratory signal had significantly high correlation with ventilatory parameters. These results support tracheal sound analysis as a simpler, robust and cost-effective modality to estimate CPET measurements.