Tesi etd-03182024-144050
Link copiato negli appunti
Tipo di tesi
Dottorato
Autore
LASSI, MICHAEL
URN
etd-03182024-144050
Titolo
Identification and evaluation of EEG markers for the diagnosis and prognosis of neural clinical conditions
Settore scientifico disciplinare
ING-INF/06
Corso di studi
Istituto di Biorobotica - PHD IN BIOROBOTICA
Commissione
relatore Dott. MAZZONI, ALBERTO
Membro Prof. ARTONI, FIORENZO
Membro Prof.ssa BESSI, VALENTINA
Membro Prof. MICERA, SILVESTRO
Membro Prof. ARTONI, FIORENZO
Membro Prof.ssa BESSI, VALENTINA
Membro Prof. MICERA, SILVESTRO
Parole chiave
- EEG
- machine learning
- Alzheimer's disease
- stroke
Data inizio appello
27/06/2024;
Disponibilità
parziale
Riassunto analitico
One century ago, Hans Berger acquired the first electroencephalographic (EEG) signal from the human brain. Since then, this was a precious instrument to non-invasively monitor neural correlates of a variety of brain conditions, ranging from motion to behavior and cognition. EEG also plays a critical role in clinical neurology, in contexts such as epileptic seizures detection and sleep disorders monitoring.
In this thesis, I present two advancements able to further broaden the clinical applications of this technology. First, I show how a rich set of signal processing tools, including machine learning and deep learning, can be used in the analysis of EEG recordings. Indeed, the use of EEG in the clinical setting has been challenging due to the difficulties in interpreting this complex multivariate signal. I show how these tools can help correctly characterize clinical EEG signals, by detecting the most salient patterns defining pathological conditions. Second, these tools can extend the use of EEG recordings from cross-sectional monitoring of these conditions to longitudinal prediction of future evolution of patient’s status, both in terms of disease progression and recovery associated with rehabilitation.
This is carried out in two different specific applications: the search for biomarkers of pre-clinical stages of Alzheimer’s disease and the search for biomarkers of post-stroke upper limb motor recovery.
Alzheimer’s disease (AD) is the most common form of dementia and identifying it at its earliest pre-clinical stages would be pivotal to deliver the best possible clinical care. For this reason, I addressed the classification of subjective cognitive decline and mild cognitive impairment, two conditions that might evolve in Alzheimer’s dementia. The works included in the thesis show that EEG can give a substantial contribution to the diagnosis of these states and possibly assess the risk of disease progression. This is of relevance, as EEG is a cheap and non-invasive tool when compared to current gold standard procedures to extract biomarkers of dementia (e.g. cerebrospinal fluid biopsy, PET imaging).
Stroke is a leading cause of disabilities worldwide. For every therapeutical and rehabilitative path, recovery post-stroke is determined by a variety of factors, such as patient’s status, lesion volume and location. Hence, defining a personalized assessment of post-stroke recovery potential is essential to propose patient-tailored rehabilitation protocols. I addressed the identification of EEG biomarkers that could track and predict the motor recovery of the upper limb after the stroke event. I investigated neural correlates of motor recovery both in the case of conventional therapy and in the case of robot-assisted rehabilitation. The proposed biomarkers are effective in determining the likelihood of positive rehabilitative outcome for each patient.
The two projects also share two critical features. Firstly, a key aspect to rehabilitation lies in understanding how the lesion induces overall network modifications. Similarly, AD is characterized by a systemic network failure rather than a specific dysfunction. Therefore, our network analysis tools are beneficial for both applications.
Secondly, in both cases, the focus is not on a static snapshot of the network's condition, but on its evolution due to synaptic plasticity and neurodegeneration processes. This justifies the need for machine learning approaches to predict network evolution in both scenarios.
In the two applications, the use of these tools paved the way for further analyses integrating the EEG signal with multimodal assessments. The proposed tools establish the ground for data-driven methodologies that will help better understand the pathophysiology of neurological conditions and hence support clinical decision-making.
In this thesis, I present two advancements able to further broaden the clinical applications of this technology. First, I show how a rich set of signal processing tools, including machine learning and deep learning, can be used in the analysis of EEG recordings. Indeed, the use of EEG in the clinical setting has been challenging due to the difficulties in interpreting this complex multivariate signal. I show how these tools can help correctly characterize clinical EEG signals, by detecting the most salient patterns defining pathological conditions. Second, these tools can extend the use of EEG recordings from cross-sectional monitoring of these conditions to longitudinal prediction of future evolution of patient’s status, both in terms of disease progression and recovery associated with rehabilitation.
This is carried out in two different specific applications: the search for biomarkers of pre-clinical stages of Alzheimer’s disease and the search for biomarkers of post-stroke upper limb motor recovery.
Alzheimer’s disease (AD) is the most common form of dementia and identifying it at its earliest pre-clinical stages would be pivotal to deliver the best possible clinical care. For this reason, I addressed the classification of subjective cognitive decline and mild cognitive impairment, two conditions that might evolve in Alzheimer’s dementia. The works included in the thesis show that EEG can give a substantial contribution to the diagnosis of these states and possibly assess the risk of disease progression. This is of relevance, as EEG is a cheap and non-invasive tool when compared to current gold standard procedures to extract biomarkers of dementia (e.g. cerebrospinal fluid biopsy, PET imaging).
Stroke is a leading cause of disabilities worldwide. For every therapeutical and rehabilitative path, recovery post-stroke is determined by a variety of factors, such as patient’s status, lesion volume and location. Hence, defining a personalized assessment of post-stroke recovery potential is essential to propose patient-tailored rehabilitation protocols. I addressed the identification of EEG biomarkers that could track and predict the motor recovery of the upper limb after the stroke event. I investigated neural correlates of motor recovery both in the case of conventional therapy and in the case of robot-assisted rehabilitation. The proposed biomarkers are effective in determining the likelihood of positive rehabilitative outcome for each patient.
The two projects also share two critical features. Firstly, a key aspect to rehabilitation lies in understanding how the lesion induces overall network modifications. Similarly, AD is characterized by a systemic network failure rather than a specific dysfunction. Therefore, our network analysis tools are beneficial for both applications.
Secondly, in both cases, the focus is not on a static snapshot of the network's condition, but on its evolution due to synaptic plasticity and neurodegeneration processes. This justifies the need for machine learning approaches to predict network evolution in both scenarios.
In the two applications, the use of these tools paved the way for further analyses integrating the EEG signal with multimodal assessments. The proposed tools establish the ground for data-driven methodologies that will help better understand the pathophysiology of neurological conditions and hence support clinical decision-making.
File
| Nome file | Dimensione |
|---|---|
Ci sono 1 file riservati su richiesta dell'autore. |
|