Symbolic dynamics to enhance diagnostic ability of portable oximetry from the Phone Oximeter in the detection of paediatric sleep apnoea

Abstract

Symbolic dynamics provides complementary information to conventional oximetry analysis enabling reliable detection of moderate-to-severe paediatric OSAHS from portable oximetry.

Convolutional Neural Networks to Detect Pediatric Apnea-Hypopnea Events from Oximetry

“…The use of the oximetry signal to assist in the diagnosis of pediatric SAHS has been frequently advocated, being especially suitable for children [3]. The development of portable monitoring devices has allowed to record the SpO2 signals with a pulse oximeter at the patient's home in children suffering from SAHS [10].…”

“…Recent works have focused on the application of signal processing techniques to the SpO2 signal in order to assist in the diagnosis of pediatric SAHS [5]- [10]. These studies used conventional machine-learning approaches: feature extraction, feature selection, and classification and regression algorithms.…”

“…These studies used conventional machine-learning approaches: feature extraction, feature selection, and classification and regression algorithms. These studies reported accuracies in the range 75-85%, 81-84%, and 85-91% for the detection of mild (AHI≥1), moderate (AHI≥5), and severe SAHS (AHI≥10), respectively [5]- [10]. However, these studies focused on detecting the presence and severity of pediatric SAHS [5]- [10], whereas our study aims at detecting AH events.…”

“…SpO2 value usually presents drops associated to AH events [2], thus being useful to detect these episodes. In this sense, previous works have employed a methodology based on the automated analysis of the SpO2 signal in order to assist in the diagnosis of childhood SAHS [5]- [10].…”

“…These studies have applied conventional oximetric indices, as well as time, frequency, and nonlinear methods to obtain features that may characterize the properties of the SpO2 signal [5]- [10]. In addition, feature selection algorithms were used to reduce the dimensionality and improve performance, whereas classification and regression algorithms were employed to detect pediatric SAHS [5]- [10]. However, these machinelearning based frameworks require to determine which features to extract from the physiological signal, as well as to decide how many features are needed, being a difficult task [11].…”