Tesi etd-04082024-112052
Link copiato negli appunti
Tipo di tesi
Dottorato
Autore
LIUZZI, PIERGIUSEPPE
URN
etd-04082024-112052
Titolo
Patient stratification and prognosis of consciousness disorders: theory-based markers and machine learning models
Settore scientifico disciplinare
ING-IND/34
Corso di studi
Istituto di Biorobotica - PHD IN BIOROBOTICA
Commissione
relatore Dott. MANNINI, ANDREA
Membro Prof. ODDO, CALOGERO MARIA
Membro Prof. MAZZONI, ALBERTO
Presidente Prof.ssa CARROZZA, MARIA CHIARA
Membro Prof. ODDO, CALOGERO MARIA
Membro Prof. MAZZONI, ALBERTO
Presidente Prof.ssa CARROZZA, MARIA CHIARA
Parole chiave
- Disorders of Consciousness
- Machine Learning; EEG; ECG; Neurorehabilitation
Data inizio appello
24/10/2024;
Disponibilità
parziale
Riassunto analitico
Consciousness, a fundamental aspect of human experience, remains the most intriguing phenomenon in neuroscience. This doctoral thesis delves into the exploration of markers of consciousness through a multidisciplinary lens drawing from neuroscience, engineering, and mathematics. It starts with a review of the existing literature, attempting to analyze systematically the EEG processing strategies and the use of machine learning for brain injured patients prognosis estimation. Building upon this foundation, the research methodology integrates electrophysiological recordings (central and peripheral), behavioral assessments, and functional evaluations to investigate consciousness from multiple perspectives. Through meticulous experimental design and analysis, this thesis examines neural oscillations, connectivity patterns, and information processing dynamics associated with conscious awareness across different modalities/tasks.
Hub-like behavior, integration, and efficiency were found to change at different consciousness levels as measured by clinical scales in patients with a disorder of consciousness. Graph- and fractal-like determinants from low-density EEG recordings from clinical routine were able to detect patients in a minimally conscious state +/-, with a focus on language-related areas. Microstate analysis revealed how the transition from left-right to antero-posterior topographies is an independent predictor of consciousness recovery which is more informative than the admission consciousness level. Also, a higher sequence complexity and the presence of frontal-like topography were found in groups of patients with alpha-dominant background, cortical reactivity, and antero-posterior gradient. Including the autonomic system (cardiac and respiratory variability) enabled automated diagnostic pipelines that outperformed in accuracy only EEG and/or clinical data. Investigating the brain-heart axis showed proportionality between the difference in information content between EEG and ECG and consciousness, and it identified hemorrhagic lesions. Lastly, attempting to substitute magnetic stimulation with an endogenous stimulus, the heartbeat evoked potential, opened a new, less-invasive, line of inquiry to observe causal brain desynchronization in disorders of consciousness. From a translational point of view, some of the aforementioned markers of consciousness converged into interpretable decision support tools embedded with graphical interfaces, easily translatable to clinical practice.
Overall, besides reporting pivotal advancements into the central and peripheral neurophysiology of consciousness, the presented results have the potential to improve clinical practice by decreasing uncertainty in the most complex neurological condition and by introducing multi-domain explanations of the models' predictions into the clinicians' prognosis.
Hub-like behavior, integration, and efficiency were found to change at different consciousness levels as measured by clinical scales in patients with a disorder of consciousness. Graph- and fractal-like determinants from low-density EEG recordings from clinical routine were able to detect patients in a minimally conscious state +/-, with a focus on language-related areas. Microstate analysis revealed how the transition from left-right to antero-posterior topographies is an independent predictor of consciousness recovery which is more informative than the admission consciousness level. Also, a higher sequence complexity and the presence of frontal-like topography were found in groups of patients with alpha-dominant background, cortical reactivity, and antero-posterior gradient. Including the autonomic system (cardiac and respiratory variability) enabled automated diagnostic pipelines that outperformed in accuracy only EEG and/or clinical data. Investigating the brain-heart axis showed proportionality between the difference in information content between EEG and ECG and consciousness, and it identified hemorrhagic lesions. Lastly, attempting to substitute magnetic stimulation with an endogenous stimulus, the heartbeat evoked potential, opened a new, less-invasive, line of inquiry to observe causal brain desynchronization in disorders of consciousness. From a translational point of view, some of the aforementioned markers of consciousness converged into interpretable decision support tools embedded with graphical interfaces, easily translatable to clinical practice.
Overall, besides reporting pivotal advancements into the central and peripheral neurophysiology of consciousness, the presented results have the potential to improve clinical practice by decreasing uncertainty in the most complex neurological condition and by introducing multi-domain explanations of the models' predictions into the clinicians' prognosis.
File
| Nome file | Dimensione |
|---|---|
Ci sono 1 file riservati su richiesta dell'autore. |
|