0378 Airflow Estimation During Sleep Using Abdomen Movement and Breathing Sound

Abstract Introduction Accurate airflow measurement is essential in sleep monitoring. Although accelerometers paired with tracheal breath sounds have shown potential as alternatives to airflow sensors, accelerometer signals are frequently affected by artifacts from non-respiratory movements. Capacitive sensors offer a more direct and potentially reliable method for capturing respiratory motion. This study develops an algorithm for airflow estimation by combining capacitive signals from a wireless abdomen-worn sensor (Soom) with breathing sounds recorded via a microphone. Methods Ninety-seven participants underwent overnight polysomnography (PSG). Data were simultaneously collected from three sources: the Soom sensor (5 Hz), a microphone (8,000 Hz), and the nasal pressure channel (50 Hz). All obtained signals were time-synchronized. The Soom device captured capacitance and tri-axial accelerometer measurements, which were pre-processed and used as inputs for our previously established apnea detection model. A sound envelope was derived from the audio recording using the short-time Fourier transform. The sound envelope and nasal pressure were resampled to 5 Hz. Synchronized signals were segmented into one-minute intervals, normalized, and labeled as normal (no apneas) or abnormal ( ≥10 seconds of apneic events). Airflow was estimated from capacitance using a linear regression model and, during abnormal periods, further refined using the sound envelope to better capture airflow reduction. The estimated airflow was evaluated against nasal pressure using repeated-measure correlation (r) and mean squared error (MSE). Results The proposed algorithm demonstrated moderate performance in airflow estimation, with r = 0.74 ± 0.17 and MSE = 0.45 ± 0.31. In normal segments, r and MSE were 0.8 ± 0.15 and 0.42 ± 0.29, respectively. In abnormal events, these values were 0.65 ± 0.15 and 0.63 ± 0.29, respectively. Conclusion This study presented an effective algorithm that combines capacitive and sound data to estimate airflow, potentially enabling improved respiratory event classification in home sleep apnea testing without the need for nasal pressure. Support (if any) This work was supported by Innovative Human Resource Development for Local Intellectualization program through the Institute of Information & Communications Technology Planning & Evaluation (IITP) grant funded by the Korea government (MSIT)(IITP-2025-RS-2022-00156361), and the Technology Innovation Program (RS-2023-00236657) funded by the Ministry of Trade Industry & Energy (MOTIE, Korea).