This thesis investigates the use of electromagnetic waves for the prediction of osteoporosis, a condition characterized by the progressive deterioration of bone microarchitecture and associated with significant clinical and socioeconomic consequences. The aim of this work is to develop and evaluate Machine Learning models capable of leveraging data derived from such signals to support early screening activities. Existing literature highlights the effectiveness of established diagnostic techniques, such as bone densitometry, while also emphasizing the need for more accessible and cost-effective alternative solutions. In this context, several studies have proposed the use of portable devices based on electromagnetic waves; however, challenges remain regarding model robustness and their transferability to real-world scenarios. Within this framework, the present work contributes to this line of research by proposing an analysis based on Machine Learning models trained on datasets obtained through electromagnetic waves, with the goal of assessing their performance under more realistic conditions of use. The results show a limited correlation between the considered variables and the presence of the disease; nevertheless, evaluation metrics indicate a significant predictive capability of the developed models. This suggests that, even in the presence of non-strongly linear relationships, such approaches can effectively contribute to screening processes. Furthermore, patient profiling emerges as a relevant factor, and its integration—together with the availability of larger datasets—could further improve model performance. Overall, the findings indicate that low-cost technologies combined with Machine Learning techniques may represent a valuable tool for the prevention and early detection of osteoporosis. However, further validation on larger datasets and real-world populations is required to enhance reliability and support clinical adoption.
Questa tesi analizza l’utilizzo di onde elettromagnetiche per la predizione dell’osteoporosi, una patologia caratterizzata dal progressivo deterioramento della microarchitettura ossea e associata a rilevanti conseguenze cliniche e socioeconomiche. L’obiettivo del lavoro è sviluppare e valutare modelli di Machine Learning in grado di sfruttare dati derivati da tali segnali per supportare attività di screening precoce. La letteratura evidenzia l’efficacia di tecniche diagnostiche consolidate, come la densitometria ossea, ma anche la necessità di soluzioni alternative più accessibili ed economiche. In questo contesto, diversi studi propongono l’impiego di dispositivi portatili basati su onde elettromagnetiche, pur evidenziando criticità legate alla robustezza dei modelli e alla loro trasferibilità in scenari reali. Il presente lavoro si inserisce in tale ambito, proponendo un’analisi basata su modelli di Machine Learning addestrati su dataset ottenuti tramite onde elettromagnetiche, con l’obiettivo di valutarne le prestazioni in condizioni più realistiche di utilizzo. I risultati mostrano una correlazione limitata tra le variabili considerate e la presenza della patologia; tuttavia, le metriche di valutazione evidenziano una capacità predittiva significativa. Ciò suggerisce che, anche in presenza di relazioni non fortemente lineari, tali approcci possano contribuire efficacemente allo screening. Inoltre, emerge il ruolo rilevante della profilazione del paziente, la cui integrazione, insieme alla disponibilità di dataset più ampi, potrebbe migliorare ulteriormente le prestazioni dei modelli. Nel complesso, i risultati indicano che tecnologie a basso costo integrate con tecniche di Machine Learning possono rappresentare un valido supporto per la prevenzione e l’identificazione precoce dell’osteoporosi. Tuttavia, sono necessarie ulteriori validazioni su dataset più estesi e popolazioni reali per migliorarne l’affidabilità e favorirne l’applicazione clinica.
Machine Learning per la predizione dell'osteoporosi: analisi delle onde elettromagnetiche come surrogato dei raggi X.
ZANATTA, GIORGIO
2025/2026
Abstract
This thesis investigates the use of electromagnetic waves for the prediction of osteoporosis, a condition characterized by the progressive deterioration of bone microarchitecture and associated with significant clinical and socioeconomic consequences. The aim of this work is to develop and evaluate Machine Learning models capable of leveraging data derived from such signals to support early screening activities. Existing literature highlights the effectiveness of established diagnostic techniques, such as bone densitometry, while also emphasizing the need for more accessible and cost-effective alternative solutions. In this context, several studies have proposed the use of portable devices based on electromagnetic waves; however, challenges remain regarding model robustness and their transferability to real-world scenarios. Within this framework, the present work contributes to this line of research by proposing an analysis based on Machine Learning models trained on datasets obtained through electromagnetic waves, with the goal of assessing their performance under more realistic conditions of use. The results show a limited correlation between the considered variables and the presence of the disease; nevertheless, evaluation metrics indicate a significant predictive capability of the developed models. This suggests that, even in the presence of non-strongly linear relationships, such approaches can effectively contribute to screening processes. Furthermore, patient profiling emerges as a relevant factor, and its integration—together with the availability of larger datasets—could further improve model performance. Overall, the findings indicate that low-cost technologies combined with Machine Learning techniques may represent a valuable tool for the prevention and early detection of osteoporosis. However, further validation on larger datasets and real-world populations is required to enhance reliability and support clinical adoption.| File | Dimensione | Formato | |
|---|---|---|---|
|
Zanatta_Giorgio.pdf
Accesso riservato
Dimensione
8.33 MB
Formato
Adobe PDF
|
8.33 MB | Adobe PDF |
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/20.500.12608/107600