On-Line Detection of Apnea/Hypopnea Events Using SpO $_{\bf 2}$ Signal: A Rule-Based Approach Employing Binary Classifier Models

摘要: This paper presents an online method for automatic detection of apnea/hypopnea events, with the help oxygen saturation (SpO2) signal, measured at fingertip by Bluetooth nocturnal pulse oximeter. Event is performed identifying abnormal data segments from recorded SpO 2 employing a binary classifier model based on support vector machine (SVM). Thereafter segment further analyzed to detect different states within segment, i.e., steady, desaturation, and resaturation, another SVM-based ensemble model. Finally, heuristically obtained rule-based system used identify events time-sequenced decisions these models. In developmental phase, set 34 time domain-based features was extracted segmented SpO2 signal using overlapped windowing technique. Later, optimal selected basis recursive feature elimination A total subjects were included in study. The results show average event accuracies 96.7% 93.8% offline tests, respectively. proposed provides direct estimation index relatively inexpensive widely available Moreover, can be monitored accessed physicians through LAN/WAN/Internet extended deploy Bluetooth-enabled mobile phones.