This study delves into the denoising of Electroencephalogram (EEG) signals, which are often contaminated by Electromyography (EMG) and Electrooculography (EOG) noise, using an autoencoder model. Recognizing the critical role of EEG signals in understanding brain function and activity, this research focuses on developing an effective denoising approach to improve signal quality and reliability. The autoencoder model, a type of neural network, is trained to reconstruct clean EEG signals from noisy input data, thereby effectively removing EMG and EOG artifacts. Various versions of the experiments, incorporating different architectures and dataset constructions, are explored to optimize denoising performance. The effectiveness of each model is evaluated based on its ability to reduce noise while preserving essential EEG features. Through extensive experimentation and analysis, the study elucidates the efficacy of deep learning techniques in mitigating noise artifacts and improve the reliability of EEG recordings. The findings not only contribute to advancing signal processing methodologies in neuroscience but also hold significant implications for clinical applications, such as the diagnosis and monitoring of neurological disorders. By refining denoising approaches and leveraging the power of deep learning, researchers can unlock the full potential of EEG as a non-invasive tool for studying brain function and activity.
Questo studio si occupa di pulire i segnali elettroencefalografici (EEG), spesso contaminati dal rumore dell’elettromiografia (EMG) e dell’elettrooculografia (EOG), utilizzando un’architettura di deep learning. Riconoscendo il ruolo critico dei segnali EEG nella comprensione della funzione e dell’attività cerebrale, questa ricerca si concentra sullo sviluppo di un approccio di denoising efficace per migliorare la qualità e l’affidabilità del segnale. Il modello autoencoder, un tipo di rete neurale, viene addestrato per ricostruire segnali EEG puliti da dati di ingresso rumorosi, rimuovendo così efficacemente gli artefatti EMG ed EOG. Per ottimizzare le prestazioni di denoising vengono eseguiti diversi esperimenti, che incorporano diverse architetture e costruzioni di set di dati. L’efficacia di ogni modello viene valutata in base alla sua capacità di ridurre il rumore preservando le caratteristiche EEG essenziali. Attraverso un’ampia sperimentazione e analisi, lo studio chiarisce l’efficacia delle tecniche di deep learning nel mitigare gli artefatti da rumore e migliorare l’affidabilità delle registrazioni EEG. I risultati non solo contribuiscono all’avanzamento delle metodologie di elaborazione del segnale nelle neuroscienze, ma hanno anche implicazioni significative per le applicazioni cliniche, come la diagnosi e il monitoraggio dei disturbi neurologici. Affinando gli approcci di denoising e sfruttando la potenza del deep learning, i ricercatori possono sbloccare il pieno potenziale dell’EEG come strumento non invasivo per lo studio della funzione e dell’attività cerebrale.
A deep learning architecture for the denoising of the electroencephalogram signals affected by electromyography and electrooculography noise
Barone, Annarita
2023/2024
Abstract
This study delves into the denoising of Electroencephalogram (EEG) signals, which are often contaminated by Electromyography (EMG) and Electrooculography (EOG) noise, using an autoencoder model. Recognizing the critical role of EEG signals in understanding brain function and activity, this research focuses on developing an effective denoising approach to improve signal quality and reliability. The autoencoder model, a type of neural network, is trained to reconstruct clean EEG signals from noisy input data, thereby effectively removing EMG and EOG artifacts. Various versions of the experiments, incorporating different architectures and dataset constructions, are explored to optimize denoising performance. The effectiveness of each model is evaluated based on its ability to reduce noise while preserving essential EEG features. Through extensive experimentation and analysis, the study elucidates the efficacy of deep learning techniques in mitigating noise artifacts and improve the reliability of EEG recordings. The findings not only contribute to advancing signal processing methodologies in neuroscience but also hold significant implications for clinical applications, such as the diagnosis and monitoring of neurological disorders. By refining denoising approaches and leveraging the power of deep learning, researchers can unlock the full potential of EEG as a non-invasive tool for studying brain function and activity.File | Dimensione | Formato | |
---|---|---|---|
TesiAnnarita.pdf
accessibile in internet per tutti
Descrizione: Thesis
Dimensione
9.89 MB
Formato
Adobe PDF
|
9.89 MB | Adobe PDF | Visualizza/Apri |
Executive_Summary___Annarita_Barone.pdf
accessibile in internet per tutti
Descrizione: Executive Summary
Dimensione
848.76 kB
Formato
Adobe PDF
|
848.76 kB | Adobe PDF | Visualizza/Apri |
I documenti in POLITesi sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
https://hdl.handle.net/10589/218035