This study was performed at the NICU of the University Hospitals of Leuven, Belgium and approved by the Ethics Committee of the University Hospitals of Leuven, Belgium. Neonates were enrolled in the study after informed parental consent. The dataset consisted of 26 preterm neonates with gestational age ≤32 weeks. Neonates were retrospectively selected as ‘normal’, based on strict inclusion criteria: (1) A normal neurodevelopmental outcome score at 9 and 24 months corrected age (Bayley Scales of Infant Development-II, mental and motor function >85), (2) no use of any sedative or anti-epileptic medication during EEG registration, and (3) the absence of a severe cerebral lesion (normal cerebral ultrasonography or intraventricular hemorrhage grade ≤ II, no periventricular leukomalacia or ventricular dilatation >p97).
EEG recordings were obtained from the neonates between the first and the third week of life, followed by one recording every 2 to 3 weeks up to transfer or discharge. This resulted in 89 recordings ranging from 27 to 42 weeks PMA. The age distribution of this dataset is presented in the histogram ofFig. 1 .
Mean EEG monitoring time was 4 h 55 min (range 1 h 40 min–9 h 00 min), in accordance with neonatal EEG surveillance guidelines28 (link) to acquire at least two complete sleep cycles. Feeding and care were carried out per the normal routine of the NICU. Kangaroo Care was encouraged and allowed during the recordings as part of the application of the Newborn Individualized Developmental Care and Assessment Program. All EEG recordings were recorded with nine electrodes (Fp1, Fp2, C3, C4, T3, T4, O1, O2, and reference electrode Cz) placed per the modified international 10–20 standard locations (BRAIN RT, OSG equipment, Mechelen, Belgium) at a sampling frequency of 250 Hz. In premature infants <36 weeks PMA, unobtrusive sleep EEG monitoring was performed including a channel for respiratory activity, electrocardiogram and oxygen saturation. Infants ≥36 weeks PMA had an overnight PSG recording with 12-channel EEG, electrocardiogram, oxygen saturation, electromyogram, 2 electro-oculograms, piezoelectric belts (to measure abdominal and thoracic respiratory effort), and a nasal thermistor (for airflow monitoring before discharge).
In the remainder of this paper, the first 34 visually labeled recordings that were obtained for algorithm development and optimization, are referred to as the training set. The subsequent 55 labeled recordings obtained were referred to as the test set, used solely to assess final algorithm performance.
EEG recordings were obtained from the neonates between the first and the third week of life, followed by one recording every 2 to 3 weeks up to transfer or discharge. This resulted in 89 recordings ranging from 27 to 42 weeks PMA. The age distribution of this dataset is presented in the histogram of
Mean EEG monitoring time was 4 h 55 min (range 1 h 40 min–9 h 00 min), in accordance with neonatal EEG surveillance guidelines28 (link) to acquire at least two complete sleep cycles. Feeding and care were carried out per the normal routine of the NICU. Kangaroo Care was encouraged and allowed during the recordings as part of the application of the Newborn Individualized Developmental Care and Assessment Program. All EEG recordings were recorded with nine electrodes (Fp1, Fp2, C3, C4, T3, T4, O1, O2, and reference electrode Cz) placed per the modified international 10–20 standard locations (BRAIN RT, OSG equipment, Mechelen, Belgium) at a sampling frequency of 250 Hz. In premature infants <36 weeks PMA, unobtrusive sleep EEG monitoring was performed including a channel for respiratory activity, electrocardiogram and oxygen saturation. Infants ≥36 weeks PMA had an overnight PSG recording with 12-channel EEG, electrocardiogram, oxygen saturation, electromyogram, 2 electro-oculograms, piezoelectric belts (to measure abdominal and thoracic respiratory effort), and a nasal thermistor (for airflow monitoring before discharge).
In the remainder of this paper, the first 34 visually labeled recordings that were obtained for algorithm development and optimization, are referred to as the training set. The subsequent 55 labeled recordings obtained were referred to as the test set, used solely to assess final algorithm performance.
Free full text: Click here
