DI-UMONS : Dépôt institutionnel de l’université de Mons

Recherche transversale
(titres de publication, de périodique et noms de colloque inclus)
2019-02-05 - Colloque/Présentation - poster - Anglais - page(s)

Itani Sarah , Thanou Dorina, Rossignol Mandy , Lecron Fabian , Fortemps Philippe , "Exploiting the Structure-Function Interplay of Brain Regions for the Diagnosis of Autism Spectrum Disorder: A Machine Learning Approach" in The Childbrain Conference , Leuven, Belgique, 2019

  • Codes CREF : Intelligence artificielle (DI1180), Modèles mathématiques d'aide à la décision (DI1151), Informatique médicale (DI3314), Informatique mathématique (DI1160), Ingénierie biomédicale (DI3900)
  • Unités de recherche UMONS : Psychologie cognitive et Neuropsychologie (P325), Management de l'Innovation Technologique (F113), Mathématique et Recherche opérationnelle (F151)
  • Instituts UMONS : Institut de Recherche en Technologies de l’Information et Sciences de l’Informatique (InforTech), Institut des Sciences et Technologies de la Santé (Santé)
  • Centres UMONS : Centre de recherche interdisciplinaire en Psychophysiologie et Electrophysiologie de la cognition (CIPsE)

Abstract(s) :

(Anglais) Autism Spectrum Disorder (ASD) is a neurodevelopmental disorder which is characterized by recurring behavioral patterns, and difficulties with communication and social ability. While the prevalence of ASD is increasing, research towards the definition of a common etiology is still ongoing. Modern machine learning and network science techniques give grounds for hope of capturing the complexity of the neuropathology through the detection of explanatory markers, and the development of systems able to recommend a diagnosis. In this work, we tackle the problem of classification of neurotypical and ASD subjects by combining domain knowledge about the anatomy of the brain and the information provided by the functional Magnetic Resonance Imaging (fMRI) signals of brain regions. In particular, we model the structure of the brain as a graph, whose nodes are the brain regions, and the time-varying fMRI signals as values that reside on the nodes of that graph. We then exploit the emerging field of Graph Signal Processing to define features based on the frequency content of these signals. In order to make these features highly discriminative, we apply an extension of the Fukunaga-Koontz transform. Finally, these new markers are given as an input to a decision tree, an easily interpretable machine learning model which establishes a sequence of simple questions in order to classify the subjects. The proposed methodology is used to classify the data derived from the publicly available ABIDE (Autism Brain Imaging Data Exchange) dataset. In particular, we focus on a population of neurotypical and ASD subjects (n = 452), aged less than 18 years old. For each subject, the dataset contains Blood Level Oxygen Dependent (BOLD) signals, at resting-state, given a brain parcellation in 90 brain regions. Interestingly, the resulting decision tree achieves a classification accuracy of 75%, and outperforms state-of-the-art methods. Moreover, the analysis of the predictive markers reveals the influence of the fronto-temporal areas in the prediction of ASD, which is in line with previous findings in the literature of neuroscience. These findings indicate that exploiting jointly structural (brain topology) and functional (fMRI-based activity) information of the brain regions is more informative than focusing on each of them separately. We believe that such an approach can pave the way for a better understanding of the disease and thus the exploration of new therapeutic approaches.

Mots-clés :
  • (Anglais) diagnosis aid
  • (Anglais) artificial intelligence
  • (Anglais) autism