A Novel Approach In The Detection Of Obstructive Sleep Apnea From Electrocardiogram Signals Using Neural Network Classification Of Textural Features Extracted From Time-frequency Plots

Sleep-Disordered Breathing (SDB) is estimated to have a prevalence of 5% in middle-aged population. The population is widely thought to be under diagnosed, since the present method to detect and diagnose SDB, Nocturnal Polysomnography (NPSG), is still expensive and not accessible by most. SDB has been shown to affect the productivity and degree of life of the patient, and to have a high correlation with obesity and cognitive heart failure (CHF). Cheap and accessible means to screen the population for SDB are greatly pursued. This work presents an automatic algorithm to detect obstructive sleep apnea (OSA) events in 15-minute clips. Data is collected from 12 normal subjects (6 males, 6 females; age 46.27±9.79 years, AHI 3.82±3.25) and 14 apneic subjects (8 males, 6 females; age 49.08±8.82 years; AHI 31.21±23.90). The algorithm uses textural features extracted from co-occurrence matrices of gray-level encoded images generated by short-time discrete Fourier transform (STDFT) of the heart rate variability (HRV). Seventeen selected features are used as inputs to a 3-layer multilayer perceptron (MLP), with 45 hidden units and 4200 training epochs. A 1000-run Monte-Carlo simulation of the algorithm gave the following results: mean training sensitivity, specificity and accuracy of 99.00%, 93.42%, and 96.42%, respectively. The mean testing sensitivity, specificity and accuracy are 94.42%, 85.40%, and 90.16%, respectively.