합성곱 신경망(CNN) 기반 새로운 호흡 환기량 추정 모델 개발

Abstract

Recently, ventilation (VE) has been studied as an alternative to estimating energy expenditure. Wearable sensors used for respiratory monitoring such as VE can be affected by motion artifacts, leading to signal distortion. Therefore, this study aims to monitor respiration using a microphone sensor to estimate the respiratory parameter, VE (ventilation). A CNN model was implemented to estimate ventilation using respiratory sounds processed into Mel-spectrograms. The experiment was conducted in a treadmill environment with a protocol involving progressively increasing speed over a total of 15 minutes, during which both respiratory sounds and VE (Truth Reference) were collected simultaneously. The results showed a Pearson correlation coefficient of 0.96 ± 0.01, R² of 0.84 ± 0.07, and MAE of 6.66 ± 2.09. These results demonstrate a high correlation between respiratory sounds and VE, suggesting the potential for estimating VE using respiratory sounds.