Brain-Computer Interfaces (BCIs) are technologies that allow patients with motor disabilities to interact with external devices only through Motor Imagery (MI). These interfaces bridge the user and the system, translating neural responses into control signals that the device can employ to execute movements. This thesis work focuses on the acquisition and processing of electroencephalographic (EEG) signals recorded from five healthy subjects who were asked to perform motor imagery of the right foot, left foot, and a state of rest (Rest), with the ultimate goal of leveraging these signals to control assistive devices. The analysis of neural correlates centered on developing a pipeline that included a pre-processing phase for artifact removal, followed by feature extraction through time-frequency analysis and the calculation of ERD/ERS (Event-Related Desynchronization/Synchronization) parameters to visualize the spectrogram. Subsequently, the Logarithmic Bandpower was implemented to inspect the topographic map of cortical activation areas. Then, by calculating the Fisher Score, the most discriminative channel-frequency pairs were identified and subsequently used to train predictive models based on Linear Discriminant Analysis (LDA). The trained model was then used to predict the user's intention in order to control devices such as exoskeletons. The main challenges addressed involved defining an efficient model capable of clearly distinguishing between movement and rest classes. To cope with the partial overlap of signals caused by the lack of cortical lateralization of the feet, both models trained on single-limb data and a combined, generalized model were investigated, the latter being optimal for ensuring stability and safety during gait activation.
Le interfacce cervello computer (BCI) sono tecnologie che permettono a pazienti con disabilità motorie di interagire con dispositivi esterni tramite la sola immaginazione motoria (Motor Imagery - MI). Queste interfacce si pongono tra l'utente e il sistema con lo scopo di riconoscere le risposte celebrali traducendole in segnali che il dispositivo può riconoscere e sfruttare per eseguire il movimento. Il presente lavoro di tesi si focalizza sull'acquisizione e l'elaborazione di segnali elettroencefalografici (EEG) registrati su cinque soggetti sani a cui veniva chiesto di effettuare l'immaginazione motoria della gamba destra ,della gamba sinistra e stato di riposo (Rest) per poi sfruttarli per il controllo di dispositivi di assistenza. L'analisi dei correlati neurali si concentrava sullo sviluppo di una pipeline che comprendeva una fase di pre-processing per la rimozioni degli artefatti e la successiva estrazione delle feature tramite analisi tempo- frequenza e calcolo dei parametri ERD/ERS ( Event Related Desynchronization/Synchronization) per la visualizzazione dello spettrogramma. Successivamente è stata implementata la Logarithmic Bandpower per ispezionare la mappa topografica delle aree di attivazione corticale. In seguito, attraverso il calcolo del Fisher Score direzionato sono state identificate le coppie canale-frequenza più discriminanti, utilizzate poi per addestrare modelli predittivi basati su Analisi Discriminante Lineare (LDA). Il modello allenato è stato poi utilizzato per cercare di predire correttamente l'intenzione dell'utente per poi controllare dispositivi come gli esoscheletri. Le principali sfide affrontate hanno riguardato la definizione di un modello efficiente capace di distinguere nettamente le classi di movimento e riposo. Per fare fronte alla parziale sovrapposizione dei segnali dovuta alla mancata lateralizzazione corticale dei piedi, sono stati investigati sia modelli allenati sui dati del singolo arto sia un modello combinato e generalizzato, ottimale per garantire stabilità e sicurezza nell'attivazione del passo.
Analisi dei correlati neurali dell'immaginazione motoria per il controllo di un esoscheletro per il cammino.
PERUZZI, MARTINA
2025/2026
Abstract
Brain-Computer Interfaces (BCIs) are technologies that allow patients with motor disabilities to interact with external devices only through Motor Imagery (MI). These interfaces bridge the user and the system, translating neural responses into control signals that the device can employ to execute movements. This thesis work focuses on the acquisition and processing of electroencephalographic (EEG) signals recorded from five healthy subjects who were asked to perform motor imagery of the right foot, left foot, and a state of rest (Rest), with the ultimate goal of leveraging these signals to control assistive devices. The analysis of neural correlates centered on developing a pipeline that included a pre-processing phase for artifact removal, followed by feature extraction through time-frequency analysis and the calculation of ERD/ERS (Event-Related Desynchronization/Synchronization) parameters to visualize the spectrogram. Subsequently, the Logarithmic Bandpower was implemented to inspect the topographic map of cortical activation areas. Then, by calculating the Fisher Score, the most discriminative channel-frequency pairs were identified and subsequently used to train predictive models based on Linear Discriminant Analysis (LDA). The trained model was then used to predict the user's intention in order to control devices such as exoskeletons. The main challenges addressed involved defining an efficient model capable of clearly distinguishing between movement and rest classes. To cope with the partial overlap of signals caused by the lack of cortical lateralization of the feet, both models trained on single-limb data and a combined, generalized model were investigated, the latter being optimal for ensuring stability and safety during gait activation.| File | Dimensione | Formato | |
|---|---|---|---|
|
Peruzzi_Martina.pdf
accesso aperto
Dimensione
3.77 MB
Formato
Adobe PDF
|
3.77 MB | 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/20.500.12608/110901