Tesi etd-05152025-171302
Link copiato negli appunti
Tipo di tesi
Dottorato
Autore
CERADINI, MATTEO
URN
etd-05152025-171302
Titolo
Advancing neuroprostheses control with brain and body-machine interfaces: from non-invasive to implantable solutions
Settore scientifico disciplinare
ING-INF/06
Corso di studi
Istituto di Biorobotica - PHD IN BIOROBOTICA
Commissione
relatore Prof. MICERA, SILVESTRO
Membro Prof. Gernot Müller-Putz
Presidente Prof. LUCA TONIN
Membro Prof.ssa CYNTHIA CHESTEK
Membro Prof. Gernot Müller-Putz
Presidente Prof. LUCA TONIN
Membro Prof.ssa CYNTHIA CHESTEK
Parole chiave
- brain-computer interface
- body-machine interface
- neural decoding
- neuroprostheses
- virtual reality
- explainable AI
Data inizio appello
22/12/2025;
DisponibilitÃ
completa
Riassunto analitico
Restoring voluntary motor function after paralysis remains a central challenge in neuroprosthetics. This thesis work had the objective of advancing the field by developing and integrating brain- and body-machine interface (BCI/BMI) technologies across different non-invasive and implantable solutions, and by leveraging immersive virtual reality (VR) both for control-strategy selection and for evaluating its readiness for rehabilitation.
First, non-invasive EEG-based BCI are designed and validated to decode upper-limb movement intention, with longitudinal online studies demonstrating that subject learning improves consistently real-time decoding performance; these findings are further extended toward translation in individuals with spinal cord injury (SCI).
Second, complementary body-machine interfaces are explored using surface EMG to decode homologous grasp and finger-movement attempts in people with SCI, enabling intuitive and robust control of a virtual hand with fast-donning wearable sensors in a closed-loop setting.
Third, implanted neural and EMG recordings in non-human primates are used to evaluate sequence deep-learning models for continuous decoding. Those architectures, including transformer-based models, yield high-fidelity, real-time prediction of finger kinematics. Additionally a layer-wise relevance propagation (LRP) explainability framework is introduced to elucidate temporal and channel-wise feature relevance, improving interpretability of neural decoders.
Finally, an immersive VR protocol is developed to let people with SCI experience and compare alternative and possible neuroprosthesis control schemes, uncovering subject-specific preferences and supporting personalized device configuration. A scoping review further highlights opportunities and considerations for immersive VR deployment in upper-extremity neurorehabilitation.
Collectively, these contributions show that combining advanced decoding algorithms, user adaptation, and immersive environments could enhances BCI/BMI readiness for neuroprosthesis control. The results support a personalized, multimodal pathway to restoring motor function, paving the way for more effective and acceptable neuroprosthetic motor control solutions for individuals with motor impairments.
First, non-invasive EEG-based BCI are designed and validated to decode upper-limb movement intention, with longitudinal online studies demonstrating that subject learning improves consistently real-time decoding performance; these findings are further extended toward translation in individuals with spinal cord injury (SCI).
Second, complementary body-machine interfaces are explored using surface EMG to decode homologous grasp and finger-movement attempts in people with SCI, enabling intuitive and robust control of a virtual hand with fast-donning wearable sensors in a closed-loop setting.
Third, implanted neural and EMG recordings in non-human primates are used to evaluate sequence deep-learning models for continuous decoding. Those architectures, including transformer-based models, yield high-fidelity, real-time prediction of finger kinematics. Additionally a layer-wise relevance propagation (LRP) explainability framework is introduced to elucidate temporal and channel-wise feature relevance, improving interpretability of neural decoders.
Finally, an immersive VR protocol is developed to let people with SCI experience and compare alternative and possible neuroprosthesis control schemes, uncovering subject-specific preferences and supporting personalized device configuration. A scoping review further highlights opportunities and considerations for immersive VR deployment in upper-extremity neurorehabilitation.
Collectively, these contributions show that combining advanced decoding algorithms, user adaptation, and immersive environments could enhances BCI/BMI readiness for neuroprosthesis control. The results support a personalized, multimodal pathway to restoring motor function, paving the way for more effective and acceptable neuroprosthetic motor control solutions for individuals with motor impairments.
File
| Nome file | Dimensione |
|---|---|
| PhD_thes...adini.pdf | 26.95 Mb |
Contatta l'autore |
|