This thesis analyzes the fundamental role of bioinformatics and Machine Learning within the context of precision medicine and early screening. The objective of this work is to demonstrate how the heterogeneous nature of biomedical data requires specific computational approaches and mathematical paradigms depending on the clinical challenge addressed. To this end, four vertical case studies are presented and analyzed: the use of Multi-View Feature Selection algorithms (G-SMuRFS model) applied to genomics for the early diagnosis of Alzheimer's disease; the analysis of biological Networks and Pathways for patient stratification and the study of immune evasion mechanisms in breast cancer; the application of Computer Vision and deep neural networks for the automated classification of dermatological lesions (nevi and melanomas); and the employment of Audio Processing for the extraction of acoustic features aimed at detecting vocal micro-tremors in Parkinson's disease. In conclusion, a comparative summary of the analyzed methodologies is offered, outlining the future perspectives of computational diagnostics toward the intelligent and multimodal integration of data.
Il presente lavoro di tesi analizza il ruolo fondamentale della bioinformatica e del Machine Learning nel contesto della medicina di precisione e dello screening precoce. L'obiettivo dell'elaborato è dimostrare come la natura eterogenea del dato biomedico richieda approcci computazionali e paradigmi matematici specifici a seconda della sfida clinica affrontata. A tal fine, vengono presentati e analizzati quattro casi di studio verticali: l'utilizzo di algoritmi di Feature Selection Multi-View (modello G-SMuRFS) applicati alla genomica per la diagnosi precoce dell'Alzheimer, l'analisi di Network e Pathway biologici per la stratificazione dei pazienti e lo studio dei meccanismi di evasione immunitaria nel tumore al seno, l'applicazione della Computer Vision e delle reti neurali profonde per la classificazione automatizzata di lesioni dermatologiche (nei e melanomi), l'impiego dell'Audio Processing per l'estrazione di feature acustiche volte a intercettare i micro-tremori vocali nella malattia di Parkinson. In conclusione, viene offerta una sintesi comparativa delle metodologie analizzate, delineando le prospettive future della diagnostica computazionale verso l'integrazione intelligente e multimodale dei dati.
Metodologie bioinformatiche e modelli di Machine Learning per la diagnostica avanzata
ZARDO, FEDERICA
2025/2026
Abstract
This thesis analyzes the fundamental role of bioinformatics and Machine Learning within the context of precision medicine and early screening. The objective of this work is to demonstrate how the heterogeneous nature of biomedical data requires specific computational approaches and mathematical paradigms depending on the clinical challenge addressed. To this end, four vertical case studies are presented and analyzed: the use of Multi-View Feature Selection algorithms (G-SMuRFS model) applied to genomics for the early diagnosis of Alzheimer's disease; the analysis of biological Networks and Pathways for patient stratification and the study of immune evasion mechanisms in breast cancer; the application of Computer Vision and deep neural networks for the automated classification of dermatological lesions (nevi and melanomas); and the employment of Audio Processing for the extraction of acoustic features aimed at detecting vocal micro-tremors in Parkinson's disease. In conclusion, a comparative summary of the analyzed methodologies is offered, outlining the future perspectives of computational diagnostics toward the intelligent and multimodal integration of data.| File | Dimensione | Formato | |
|---|---|---|---|
|
Zardo_Federica.pdf
accesso aperto
Dimensione
351.04 kB
Formato
Adobe PDF
|
351.04 kB | Adobe PDF | Visualizza/Apri |
The text of this website © Università degli studi di Padova. Full Text are published under a non-exclusive license. Metadata are under a CC0 License
https://hdl.handle.net/20.500.12608/114313