Thesis Project Form

Sleep plays a fundamental role in brain maturation, and its organization is closely linked to neurodevelopmental outcomes in preterm infants. Reliable classification of sleep stages can therefore support individualized care and early detection of developmental risks. Polysomnography (PSG) is currently the gold standard for sleep monitoring, involving the simultaneous recording of multiple physiological signals, including electroencephalography (EEG). While PSG require detailed information is invasive, resource-intensive, and unsuitable for long-term monitoring in infants. Videosomnography offers a contactless alternative by recording infant behaviour during sleep, capturing indicators such as movements and eye state, without disturbing the infants. However current approaches rely on manual scoring, which is both time-consuming and subject to variability highlighting the need for automatic, scalable, and clinically feasible sleep stage classification in neonatal care.

The long-term goal of this project to develop a video-based pipeline for the characterization of the sleep stages in preterm infants within the Neonatal Intensive Care Unit (NICU). To achieve this goal, the thesis will pursue several specific aims:

Investigation of deep learning algorithms for the detection and tracking of infants in video recordings in challenging NICU environments

Identification of parameters that can quantitatively describe infant motion during sleep

Extraction of meaningful features that could support the distinction of different sleep stages

Application of supervised machine learning techniques for sleep stage classification

The thesis will provide experience in:

Computer vision techniques, including pose estimation and action recognition, for the analysis of images and video data.

Deep learning algorithms for object detection and recognition in complex visual environments

Machine Learning techniques that will allow the classification of the data

Improve the knowledge of Matlab and python