Blood oxygen saturation alarm level analysis during mechanical lung ventilation

In this paper the possibility of SpO2 signal preprocessing is studied and the influence of signal smoothing on the alarm level and duration is investigated. Dyadic wavelet transform was applied to SpO2 signal and the reconstruction for different levels using approximation part was performed, considering the time duration of wavelet and scaling functions. 68 signals from PhysioNet MIMIC II database were analyzed in the experiment. The total duration of the alarm for patients with acute myocardial infarction using the proposed algorithm was reduced from 32 min. to 4 min. (by 87%) depending on decomposition level. For groups of patients with pneumonia the decrease from 65 to 5 minutes (by 92%) was achieved. Results obtained during this study suggest that wavelet decomposition and reconstruction might be used for manipulating the sensitivity of SpO2-assisted mechanical lung ventilator to the duration of outliers below the predefined threshold.