A semi-supervised autoencoder for autism disease diagnosis

Autism spectrum disorder (ASD) is a neurological developmental disorder that typically causes impaired communication and compromised social interactions. The current clinical assessment of ASD is typically based on behavioral observations and lack of the understanding of the neurological mechanism a...

Full description

Saved in:
Bibliographic Details
Published in:Neurocomputing (Amsterdam) Vol. 483; pp. 140 - 147
Main Authors: Yin, Wutao, Li, Longhai, Wu, Fang-Xiang
Format: Journal Article
Language:English
Published: Elsevier B.V 28.04.2022
Subjects:
Artificial intelligence
Autoencoders
Brain disorders diagnosis
fMRI
Semi-supervised learning
fMRI
Brain disorders diagnosis
Semi-supervised learning
Autoencoders
Artificial intelligence
ISSN:0925-2312, 1872-8286
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Autism spectrum disorder (ASD) is a neurological developmental disorder that typically causes impaired communication and compromised social interactions. The current clinical assessment of ASD is typically based on behavioral observations and lack of the understanding of the neurological mechanism and the progression of the brain development. The functional magnetic resonance imaging (fMRI) data is one of the commonly-used imaging modalities for understanding human brain mechanisms as well as the diagnosis and treatment of brain disorders such as ASD. In this paper, we proposed a semi-supervised autoencoder (AE) for autism diagnosis using functional connectivity (FC) pattern obtained from resting-state fMRI. An unsupervised autoencoder in combination with the supervised classification networks enables semi-supervised learning in which an autoencoder for learning hidden features and a neural network based classifier are trained together. Compared to train the autoencoder and classifier in separate phases, the proposed semi-supervised learning essentially helps tune the latent feature representation learning towards the goal of classification, and thus leads to improvements in autism diagnosis performance. The proposed model is evaluated by using cross-validation methods on ABIDE I database. Experimental results demonstrate that the proposed model achieves improved classification performance, and that the proposed semi-supervised learning framework can integrate unlabelled fMRI data for better feature learning and improved classification accuracy.
ISSN:0925-2312
1872-8286
DOI:10.1016/j.neucom.2022.02.017