Neonatal Data Analysis

Sleep occupies a major portion of the lives of newborns. Sleep in newborns is significantly different than sleep in adults (both in sleep architecture and continuity). The ratio of three newborn’s behavioral states (wakefulness, active and quiet sleep) is an important indicator of the maturity of the newborn brain in clinical practice. The variability of neonatal EEG signals is related to the fast maturation of the newborn’s brain and also to the frequent changes of its behavioral states. EEG also provides useful information reflecting the function of the neonatal brain: may assist in identification of focal or generalized abnormalities, existence of potentially epileptogenic foci or ongoing seizures.

Classification of newborn PSG is a very difficult task. In contrast to the adult EEG there does not exist any publicly accessible annotated referential newborn EEG (or PSG) database. It is necessary to mention that newborn EEG evolves rather fast and there are differences between one-day and twenty-day old newborn EEG. Therefore additional information about newborn exact age is also important. There are also big differences between prematurely born babies depending on the postconceptional age. It is also important to mention that the success rate of the classification can be improved using more data from a higher number of newborns of different age.

It is difficult to make a reliable classification of all sleep stages in newborns utilizing only an EEG signal. Therefore we decided to use all available polysomnographic signals and to evaluate the information content and the contribution to improving the classification for each individual feature.

Our work

We are developing methods for differentiation between three important neonatal behavioral states: quiet sleep, active sleep and wakefulness, both in pre-term and full-term newborns. The developed algorithms are tested on real neonatal data. Obtained results can be used as a reference for developing and enhancing neonatal sleep EEG/PSG classification algorithms.

The attention is also focused on the development of appropriate visualization methods. The aim of these methods is to ease the work of medical doctors and to show trends that are not obvious when performing a manual inspection of the recorded signal.

Results

Adaptive segmentation
Fig. 1. Adaptive segmentation

artefacts detection

Fig. 2. Artefacts detection

EEG spectrogram EEG spectrogram

Fig. 3. Spectral analysis

2D EEG maps 2D EEG maps

Fig. 4. EEG mapping

Coherence analysis Coherence analysis

Fig. 5. Coherence analysis


Classification
Fig. 6. Visualization of classification results


PSGLab

PSGLab is Matlab toolbox for processing of polysomnographic (PSG) data. PSGLab implements signal preprocessing, feature extraction, classification, cluster analysis and data visualization methods.

Detailed information about PSGLab: http://bio.felk.cvut.cz/psglab/

Cooperation

Data are obtained form the cooperating medical institution - Institute for the Care of Mother and Child in Prague. All the data are scored by an experienced neurologist.