This Ph.D. thesis addresses different aspects related to neuroscience. Specifically, new methodologies are proposed to monitor and investigate the evolution of the brain network connectivity of people with neural deficits. Brain network analyses are carried out to estimate the evolution of cortical connectivity in people with Alzheimer's disease (AD) and Childhood Absence Epilepsy (CAE) in order to find possible correlations with the disorders onset. Since Electroencephalogram (EEG) is the most popular tool for investigating the cerebral electrical activity, the proposed methods are based on the analysis of EEG recordings. Indeed, abnormal patterns in the electrical potentials detected on the scalp may reflect abnormalities in the communication between neurons and can be used as diagnostic and prognostic markers. Moreover, advanced machine learning techniques are employed to build up intelligent systems to aid in diagnosis of neurological disorders. Specifically, Deep learning (DL) techniques are employed to discriminate subjects with neuropathologies by only analyzing noninvasive scalp EEG recordings. To this end, a data-driven customized Convolutional Neural Network (CNN) is proposed for differentiating subjects with Alzheimer's Disease (AD), Mild Cognitive Impairment (MCI) and Healthy Control (HC); and, a Stacked Autoencoders (SAE) architecture is proposed to learn latent features able to differentiate patients affected by psychogenic non-epileptic seizures (PNES) from HC. Experimental results show the effectiveness of DL in clinical applications, indeed, the designed DL-based systems achieved better classification performance as compared to the conventional shallow machine learning and existing state-of-the-art methods. The manuscript ends exploring the potential of DL also in other real-world applications.

Il seguente lavoro di tesi affronta diverse tematiche relative alle neuroscienze. In particolare, sono state introdotte nuove metodologie sia per lo studio longitudinale della malattia dell’Alzheimer (Alzheimer's disease, AD) sia per lo studio delle connettività elettrica cerebrale di pazienti soggetti da crisi epilettiche (crisi di assenza, Childhood Absence Epilepsy, CAE), al fine di trovare possibili correlazioni con l'insorgere dei disturbi. Poiché l’elettroencefalogramma (EEG) è lo strumento più diffuso per studiare l'attività elettrica cerebrale, i metodi proposti si basano sull'analisi delle registrazioni EEG. Infatti, comportamenti anormali nei potenziali elettrici rilevati sul cuoio capelluto possono ricondurre ad anomalie nella comunicazione tra i neuroni e possono essere quindi utilizzati come “marker” diagnostici. Inoltre, tecniche avanzate di apprendimento automatico sono state utilizzate per sviluppare sistemi intelligenti al fine di aiutare nella diagnosi dei disturbi neurologici. In particolare, l’apprendimento approfondito (noto come deep learning, DL) è stato utilizzato per discriminare soggetti con neuropatologie sulla base della registrazione EEG. A questo proposito, è stata implementata sia una rete neurale convolutiva (Convolutional Neural Network, CNN) per differenziare soggetti con Alzheimer (AD), disturbo cognitivo lievi (Mild Cognitive Impairment, MCI) e soggetti sani (Healthy Control, HC); sia, un'architettura basata su Autoencoders (Stacked Autoencoders, SAE) per imparare le caratteristiche (features) più significative in grado differenziare i pazienti con convulsioni psicogene non epilettiche (Psychogenic Non-Epileptic Seizures, PNES) dai soggetti sani. I risultati sperimentali hanno dimostrato l’efficacia del DL in ambito clinico, infatti, i sistemi proposti, basati su DL, hanno ottenuto migliori prestazioni rispetto a tecniche tradizionali e a metodi presenti in letteratura. L’elaborato termina esplorando le potenzialità del deep learning anche in altre applicazioni del mondo reale.

Brain network analysis and deep learning models for studying neurological disorders based on EEG signal processing / Ieracitano, Cosimo. - (2019 Apr 16).

Brain network analysis and deep learning models for studying neurological disorders based on EEG signal processing

IERACITANO, Cosimo
2019-04-16

Abstract

This Ph.D. thesis addresses different aspects related to neuroscience. Specifically, new methodologies are proposed to monitor and investigate the evolution of the brain network connectivity of people with neural deficits. Brain network analyses are carried out to estimate the evolution of cortical connectivity in people with Alzheimer's disease (AD) and Childhood Absence Epilepsy (CAE) in order to find possible correlations with the disorders onset. Since Electroencephalogram (EEG) is the most popular tool for investigating the cerebral electrical activity, the proposed methods are based on the analysis of EEG recordings. Indeed, abnormal patterns in the electrical potentials detected on the scalp may reflect abnormalities in the communication between neurons and can be used as diagnostic and prognostic markers. Moreover, advanced machine learning techniques are employed to build up intelligent systems to aid in diagnosis of neurological disorders. Specifically, Deep learning (DL) techniques are employed to discriminate subjects with neuropathologies by only analyzing noninvasive scalp EEG recordings. To this end, a data-driven customized Convolutional Neural Network (CNN) is proposed for differentiating subjects with Alzheimer's Disease (AD), Mild Cognitive Impairment (MCI) and Healthy Control (HC); and, a Stacked Autoencoders (SAE) architecture is proposed to learn latent features able to differentiate patients affected by psychogenic non-epileptic seizures (PNES) from HC. Experimental results show the effectiveness of DL in clinical applications, indeed, the designed DL-based systems achieved better classification performance as compared to the conventional shallow machine learning and existing state-of-the-art methods. The manuscript ends exploring the potential of DL also in other real-world applications.
16-apr-2019
Settore ING-IND/31 - ELETTROTECNICA
MORABITO, Francesco Carlo
ARENA, Felice
Doctoral Thesis
File in questo prodotto:
File Dimensione Formato  
Ieracitano Cosimo.pdf

accesso aperto

Tipologia: Tesi di dottorato
Licenza: DRM non definito
Dimensione 11.44 MB
Formato Adobe PDF
11.44 MB Adobe PDF Visualizza/Apri

I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.

Utilizza questo identificativo per citare o creare un link a questo documento: https://hdl.handle.net/20.500.12318/63625
Citazioni
  • ???jsp.display-item.citation.pmc??? ND
  • Scopus ND
  • ???jsp.display-item.citation.isi??? ND
social impact