The study of neural dysfunctions requires strong prior knowledge on brain physiology combined with expertise on data analysis, signal processing, and machine learning. One of the unsolved issues regarding epilepsy consists in the localization of pathological brain areas causing seizures. Nowadays the analysis of neural activity conducted with this goal still relies on visual inspection by clinicians and is therefore subjected to human error, possibly leading to negative surgical outcome. In absence of any evidence from standard clinical tests, medical experts resort to invasive electrophysiological recordings, such as stereoelectroencephalography to assess the pathological areas. This data is high dimensional, it could suffer from spatial and temporal correlation, as well as be affected by high variability across the population. These aspects make the automatization attempt extremely challenging. In this context, this thesis tackles the problem of characterizing drug resistant focal epilepsy. This work proposes methods to analyze the intracranial electrophysiological recordings during the interictal state, leveraging on the presurgical assessment of the pathological areas. The first contribution of the thesis consists in the design of a support tool for the identification of epileptic zones. This method relies on the multi-decomposition of the signal and similarity metrics. We built personalized models which share common usage of features across patients. The second main contribution aims at understanding if there are particular frequency bands related to the epileptic areas and if it is worthy to focus on shorter periods of time. Here we leverage on the post-surgical outcome deriving from the Engel classification. The last contribution focuses on the characterization of short patterns of activity at specific frequencies. We argue that this effort could be helpful in the clinical routine and at the same time provides useful insight for the understanding of focal epilepsy.

Machine Learning for Understanding Focal Epilepsy

D'AMARIO, VANESSA
2020-05-27

Abstract

The study of neural dysfunctions requires strong prior knowledge on brain physiology combined with expertise on data analysis, signal processing, and machine learning. One of the unsolved issues regarding epilepsy consists in the localization of pathological brain areas causing seizures. Nowadays the analysis of neural activity conducted with this goal still relies on visual inspection by clinicians and is therefore subjected to human error, possibly leading to negative surgical outcome. In absence of any evidence from standard clinical tests, medical experts resort to invasive electrophysiological recordings, such as stereoelectroencephalography to assess the pathological areas. This data is high dimensional, it could suffer from spatial and temporal correlation, as well as be affected by high variability across the population. These aspects make the automatization attempt extremely challenging. In this context, this thesis tackles the problem of characterizing drug resistant focal epilepsy. This work proposes methods to analyze the intracranial electrophysiological recordings during the interictal state, leveraging on the presurgical assessment of the pathological areas. The first contribution of the thesis consists in the design of a support tool for the identification of epileptic zones. This method relies on the multi-decomposition of the signal and similarity metrics. We built personalized models which share common usage of features across patients. The second main contribution aims at understanding if there are particular frequency bands related to the epileptic areas and if it is worthy to focus on shorter periods of time. Here we leverage on the post-surgical outcome deriving from the Engel classification. The last contribution focuses on the characterization of short patterns of activity at specific frequencies. We argue that this effort could be helpful in the clinical routine and at the same time provides useful insight for the understanding of focal epilepsy.
27-mag-2020
machine learning, epilepsy
File in questo prodotto:
File Dimensione Formato  
phdunige_3629903.pdf

accesso aperto

Tipologia: Tesi di dottorato
Dimensione 9.17 MB
Formato Adobe PDF
9.17 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/11567/1009692
Citazioni
  • ???jsp.display-item.citation.pmc??? ND
  • Scopus ND
  • ???jsp.display-item.citation.isi??? ND
social impact