Performance of deep learning algorithms in predicting autism spectrum disorders based on fMRI data

Autism spectrum disorder (ASD) is a complex neurodevelopmental disease [۱]. People with autism have unusual communication and repetitive behaviours with restricted activities [۲]. Various factors such as genetics, environment, and abnormal neural connectivity play a role in the pathogenesis of the disease [۳]. Since only the evaluation of social behaviour and language skills in an autistic patient cannot provide information about the patient's neurological patterns, using functional magnetic resonance imaging (fMRI) enables the evaluation of the brain's functional connectivity as well as obtaining precise information for neuroscientists about Autism. Deep learning algorithms due to their features such as auto extract features of the images and capturing hidden representations can be effective in the early diagnosis of Autism [۴]. The purpose of this study was to investigate the performance of deep learning algorithms in predicting ASD using fMRI data.We used scientific databases such as Google Scholar, PubMed, and Web of Science to search keywords “deep learning algorithms”, “autism spectrum disorder”, and “functional magnetic resonance imaging”. Then, we extracted the related articles and reviewed them. The obtained results indicated that various deep learning algorithms such as Conditional Generative Adversarial Network (cGAN), Artificial Neural Network (ANN), Convolutional Neural Network (CNN), and Deep Q Network (DQN) were used for ASD prediction using resting state fMRI data. Also, the accuracy and sensitivity of these approaches were determined in the range of (۶۴-۹۷%) and (۷۹-۹۰%), respectively.It can be concluded that deep learning algorithms indicate a diagnostic performance to predict ASD using resting state fMRI data.

Keywords:

“Deep learning algorithms” , “Autism spectrum disorder” , “Functional magnetic resonance imaging”