logo SBA

ETD

Digital archive of theses discussed at the University of Pisa

 

Thesis etd-04042023-015231


Thesis type
Tesi di laurea magistrale
Author
FEDERICO, GIULIO
URN
etd-04042023-015231
Thesis title
Design and development of Artificial Intelligence algorithms for the analysis of EEG signals in Autism Spectrum Disorder
Department
INGEGNERIA DELL'INFORMAZIONE
Course of study
ARTIFICIAL INTELLIGENCE AND DATA ENGINEERING
Supervisors
relatore Prof. Gennaro, Claudio
relatore Prof. Amato, Giuseppe
relatore Prof. Falchi, Fabrizio
relatore Dott.ssa Billeci, Lucia
Keywords
  • ai
  • artifacts
  • autism
  • brain
  • children
  • clustering
  • cnn
  • coherence
  • connectivity
  • crossspectraldensity
  • dnn
  • eeg
  • electroencephalogram
  • fft
  • filtering
  • Fourier
  • genetic_algorithms
  • grangercasuality
  • helmet
  • hilbert
  • ica
  • imaginarycoherency
  • ispc
  • kmeans
  • matlab
  • nsga2
  • Pareto
  • pca
  • power
  • preprocessing
  • signals
  • spectrum
  • topographic
  • trials
  • wavelet
Graduation session start date
28/04/2023
Availability
Full
Summary
Autism Spectrum Disorders (ASD) are brain development alterations with onset in the first three years of life that lead to difficulties mainly in learning, social relationships and language, as well as often repetitive behaviors. The term "spectrum" underlines how autism never presents itself in the same form but varies according to the person and the time, making it difficult to understand the boundaries with the naked eye or with more traditional methods. The main reason of this work is to further investigate these boundaries, first analyzing which characteristics to date are typical in both anatomical and functional terms in both groups, and then proceeding in this same work not only with the power distribution analysis in the bands frequency drivers over the entire scalp but also doing a connectivity analysis on each of the regions since autism is more a connectivity issue than a power issue. Through a statistical analysis and subsequently through artificial intelligence algorithms we will try to understand which brain characteristics could better distinguish one group from another, in particular we will do clustering on the original features deriving from the application of algorithms to extract power and connectivity and subsequently transforming these features into others through neural networks to understand which representations best lend themselves to this discrimination. The greater understanding of these boundaries in the pre-childhood age will help to act in a timely manner to suggest any behavioral and/or pharmacological therapies capable of greatly reducing the effects of the disorder.
File