Automatic Diagnosis of Autism Using Multilevel Wavelet Decomposition and Support Vector Machine

EasyChair Preprint 10266

Abstract

The current diagnosis of autism spectrum disorder (ASD) is very challenging due to the complex symptoms of this disease. Basically, this process is based on purely behavioral observations, which makes it a subjective method that could lead to incorrect diagnoses. To address the problem in question, in this study we propose an approach for the automatic diagnosis of autism based on Multilevel Discrete Wavelet Decomposition (MDWD) and Support Vector Machines (SVM). First, we use resting-state functional magnetic resonance imaging (rs-fMRI) from the Autism Brain Imaging Data Exchange I dataset. From these images, we extract time series of regions of interest defined by a brain atlas. Then, we apply MDWD to these time series and the resulting subseries are used for the construction of functional connectivity (FC) matrices. Finally, the FC feature vector serves as input to the SVM classifier. Our proposed method is evaluated on 175 rs-fMRI sequences. The results show that using MDWD to analyze signals provides a significant improvement in classifier performance. Our best model achieves an accuracy and F1-score of 72.5% and 63.8%, respectively.

Keyphrases: MDWD, SVM, Wavelet, asd, rs-fMRI

@booklet{EasyChair:10266,
  author    = {William Cancino and Said Pertuz},
  title     = {Automatic Diagnosis of Autism Using Multilevel Wavelet Decomposition and Support Vector Machine},
  howpublished = {EasyChair Preprint 10266},
  year      = {EasyChair, 2023}}