| KEYWORD |
Deep learning-based multi-modal sleep monitoring
keywords DEEP LEARNING, SIGNAL PROCESSING, WEARABLE DEVICES
Reference persons LUIGI BORZI'
Research Groups DAUIN - GR-24 - reSilient coMputer archItectures and LIfE Sci - SMILIES
Thesis type APPLIED, EXPERIMENTAL
Description Sleep disorders significantly affect health, and current monitoring methods like polysomnography (PSG) are costly and impractical for daily use. Wearable devices offer a non-invasive alternative but require advanced techniques to fuse multi-modal data for accurate sleep analysis. Deep learning, particularly lightweight and explainable models, can enhance the accuracy of sleep stage detection and disorder identification.
The student will analyse a comprehensive dataset of 100 subjects, monitored at night using a smartwatch. The registered physiological signals include blood volume pulse from PPG sensors, acceleration from inertial sensors, electrodermal activity, and skin temperature. Gold-standard polysomnography is available as the sleep stage reference, along with clinical information on sleep health.
The student will design, develop and evaluate a light, fast, and explainable deep learning model such as convolutional neural networks (CNNs), applied to fuse the multi-modal signals and predict sleep stages and eventual disorders. The focus will be on model accuracy, speed, and interpretability to understand which features drive predictions.
The model is expected to accurately predict sleep stages, identify key features contributing to sleep health, and offer a tool for detecting sleep disorders in a non-invasive and scalable way.
Required skills Biomedical signal processing; Matlab; Python; Artificial intelligence; Deep learning
Deadline 01/03/2025
PROPONI LA TUA CANDIDATURA