Pesola, Marisa (2025) Machine Learning and Signal Processing Strategies for Precision Medicine: Advancing biomedical data analysis for diagnosis and disease monitoring. [Tesi di dottorato]

Documento PDF Pesola_Marisa_38.pdf Visibile a [TBR] Amministratori dell'archivio Download (18MB) | Richiedi una copia

Abstract

In a world increasingly shaped by digital technologies, the potential of artificial intelligence and signal processing to improve human health is both vast and tangible. Biomedical applications of these technologies offer unprecedented opportunities to enhance diagnosis, monitoring, and therapy, ultimately contributing to better quality of life and more sustainable healthcare systems. However, their integration into clinical contexts requires methodological rigor, standardization, and a careful balance between innovation and interpretability. This doctoral thesis contributes to this evolving field by exploring the synergistic application of machine learning and advanced signal processing techniques to extract novel, clinically relevant information from biomedical data, with the overarching goal of contributing to the advancement of precision medicine. The work is structured around two main research lines, each addressing specific healthcare challenges all connected by the use of modern computational tools through a common technological framework. The first part focuses on electroencephalography (EEG) for the early detection and monitoring of neurodegenerative disorders, with emphasis on Alzheimer's disease. First, the quality and standardization of EEG recordings are addressed through artifact removal and amplitude normalization, aiming to mitigate session- and subject-specific variability and to provide robust inputs for automated analysis. Then, novel EEG-based complexity metrics are designed and evaluated, and machine learning models are employed to uncover multiscale patterns associated with disease presence and progression, %Together, these contributions enhance the reliability, comparability, and diagnostic value of EEG-derived features, thereby strengthening their potential utility in clinical assessment. The second part centers on type 1 diabetes management in the context of artificial pancreas systems. Nutritional and physiological determinants of glycemic variability are systematically examined, and data-driven predictors are developed to model and forecast glucose dynamics. Particular attention is devoted to model transparency and interpretability to ensure that the resulting insights can support informed decision-making by healthcare professionals and people living with diabetes in real-world scenarios. Across all chapters, the underlying thread is the integration of AI and signal processing to enable new forms of insight from biomedical signals, insight that would be inaccessible using traditional approaches. This technological innovation not only improves scientific understanding and clinical decision-making but also holds the potential to benefit public healthcare systems. Indeed, in the context of a doctoral scholarship dedicated to the digital transformation of public administration, this thesis discusses how data-driven solutions can support more efficient, personalized, and equitable healthcare delivery.

Downloads

Actions (login required)

Modifica documento