Amyotrophic Lateral Sclerosis (ALS) is a progressive neurodegenerative disease that severely compromises motor abilities, including both verbal and written communication, while often preserving cognitive functions. In this context, Brain-Computer Interfaces (BCIs) represent a promising technology to restore communication in patients with ALS, enabling direct interaction between the brain and external devices. Traditionally, BCIs have relied on classical machine learning algorithms, such as Linear Discriminant Analysis (LDA) and Support Vector Machines (SVM), to classify non-invasive (EEG) or intracortical neural signals. However, in recent years, the application of deep neural networks—such as Convolutional Neural Networks (CNNs), Long Short-Term Memory (LSTM), and Transformers—has shown significant potential in improving the accuracy, robustness, and generalization of neural decoding models. This thesis critically examines the role of deep learning within BCI systems, with a particular focus on communication for ALS patients. Following a clinical overview of the disease and its communicative implications, the principles of BCIs, signal acquisition techniques, and major classification challenges are introduced. A comparison between traditional methods and deep learning approaches is then presented, highlighting how the latter can more effectively capture the complex and non-linear characteristics of neural signals. Finally, through the analysis of case studies from recent scientific literature, the achieved results, remaining challenges, and future perspectives of this technology are discussed, with a view to greater clinical integration and personalized treatment.
La Sclerosi Laterale Amiotrofica (SLA) è una malattia neurodegenerativa progressiva che compromette gravemente le capacità motorie, inclusa la comunicazione verbale e scritta, pur preservando spesso le funzioni cognitive. In questo contesto, le Brain-Computer Interface (BCI) rappresentano una tecnologia promettente per ripristinare la comunicazione nei pazienti affetti da SLA, permettendo l'interazione diretta tra il cervello e dispositivi esterni. Tradizionalmente, le BCI si sono basate su algoritmi di machine learning classici, come Linear Discriminant Analysis (LDA) e Support Vector Machine (SVM), per classificare segnali neurali non invasivi (EEG) o intracorticali. Tuttavia, negli ultimi anni l’applicazione di reti neurali profonde, come Convolutional Neural Network (CNN), Long Short-Term Memory (LSTM) e Transformer, ha dimostrato un notevole potenziale nel migliorare l’accuratezza, la robustezza e la generalizzazione dei modelli di decodifica neurale. Questa tesi analizza criticamente il ruolo del deep learning all'interno dei sistemi BCI, con particolare attenzione alla comunicazione nei pazienti con SLA. Dopo una panoramica clinica della patologia e delle sue implicazioni comunicative, vengono introdotti i principi delle BCI, le tecniche di acquisizione dei segnali cerebrali e le principali sfide nella classificazione. Viene poi proposto un confronto tra metodi tradizionali e approcci di deep learning, evidenziando come quest’ultimi riescano a catturare meglio le caratteristiche complesse e non lineari dei segnali neurali. Infine, attraverso l’analisi di casi studio tratti dalla letteratura scientifica, si discutono i risultati ottenuti, le criticità ancora aperte e le prospettive future di questa tecnologia, in un’ottica di maggiore integrazione clinica e personalizzazione del trattamento.
Migliorare la comunicazione in pazienti con la SLA attraverso BCI: il ruolo del Deep Learning nella decodifica dei segnali neurali
IANNONE, ANGELICA
2024/2025
Abstract
Amyotrophic Lateral Sclerosis (ALS) is a progressive neurodegenerative disease that severely compromises motor abilities, including both verbal and written communication, while often preserving cognitive functions. In this context, Brain-Computer Interfaces (BCIs) represent a promising technology to restore communication in patients with ALS, enabling direct interaction between the brain and external devices. Traditionally, BCIs have relied on classical machine learning algorithms, such as Linear Discriminant Analysis (LDA) and Support Vector Machines (SVM), to classify non-invasive (EEG) or intracortical neural signals. However, in recent years, the application of deep neural networks—such as Convolutional Neural Networks (CNNs), Long Short-Term Memory (LSTM), and Transformers—has shown significant potential in improving the accuracy, robustness, and generalization of neural decoding models. This thesis critically examines the role of deep learning within BCI systems, with a particular focus on communication for ALS patients. Following a clinical overview of the disease and its communicative implications, the principles of BCIs, signal acquisition techniques, and major classification challenges are introduced. A comparison between traditional methods and deep learning approaches is then presented, highlighting how the latter can more effectively capture the complex and non-linear characteristics of neural signals. Finally, through the analysis of case studies from recent scientific literature, the achieved results, remaining challenges, and future perspectives of this technology are discussed, with a view to greater clinical integration and personalized treatment.| File | Dimensione | Formato | |
|---|---|---|---|
|
Tesi_Iannone.pdf
accesso aperto
Dimensione
736.2 kB
Formato
Adobe PDF
|
736.2 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/20.500.12608/97786