Fig 1.
Time portraits of (a) minute-wise annotation of the presence (1) and absence (0) of OSA as marked offline by the physician based on polysomnography (PSG) records; (b) and (c) are variations of two significant quantifiers (NPSD and LVM) of HRV signals that are sensitive to the detection of OSA.
Table 1.
Baseline characteristics of the patients from Apnea-ECG database and St. Vincent’s University Hospital/ University College Dublin Database (UCD database).
Fig 2.
Evolution pathways of the features, namely NPSD and LVM, of heart rate variability signals for a representative subject during various sleep stages.
Fig 3.
Distribution of the feature states in the Laplacian-based coordinate system (V2, V3 and V4) of a representative record.
Fig 4.
Scatter plot of prediction versus real observation of time to apnea (a) and run plot showing the predictions of the expected time to sleep apnea onset in a representative patient (b).
Fig 5.
(a) Prediction performance using randomly selected data from the testing data set for a 1–5 min-ahead prediction with a prediction horizon increment of 1 min and (b) a 1–40 min-ahead prediction with a prediction horizon increment of 5 min.
Fig 6.
Distribution of the expected time to apnea onset estimated from fT(t|x*) for 1–5 min ahead from left to right.
Fig 7.
The waveforms of NPSD, LVM, sleep stage pattern, spectral density of snoring sound, OSA annotation and one-minute ahead prediction of sleep apnea in a subject.