Attention-stacking adaptive fuzzy neural networks for autism diagnosis

The current diagnostic methods for Autism Spectrum Disorder (ASD) based on Resting-State Functional Magnetic Resonance Imaging (rs-fMRI) face two significant challenges. Firstly, the functional connectivity networks (FCNs) of the brain are subject to uncertainty due to noise interference during the...

Ausführliche Beschreibung

Gespeichert in:
Bibliographische Detailangaben
Veröffentlicht in:Proceedings (IEEE International Conference on Bioinformatics and Biomedicine) S. 5654 - 5661
Hauptverfasser: Zhang, Shuzhan, Xiao, Lei, Huang, Hui, Hu, Zhongyi
Format: Tagungsbericht
Sprache:Englisch
Veröffentlicht: IEEE 03.12.2024
Schlagworte:
Accuracy
Adaptive systems
attention
Attention mechanisms
Autism
autism spectrum disorders
Contrastive learning
domain adaptive
Fuzzy neural networks
fuzzy systems
Noise
Representation learning
Transformers
Uncertainty
ISSN:2156-1133
Online-Zugang:Volltext
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
Beschreibung
Zusammenfassung:The current diagnostic methods for Autism Spectrum Disorder (ASD) based on Resting-State Functional Magnetic Resonance Imaging (rs-fMRI) face two significant challenges. Firstly, the functional connectivity networks (FCNs) of the brain are subject to uncertainty due to noise interference during the rs-fMRI acquisition process. Secondly, variations in equipment parameters across different imaging sites lead to diverse distributions of collected features. To address these issues, this paper introduces a novel Multi-Task Stacked Adaptive Fuzzy Neural Network (AT-STTSK) specifically designed for Autism Spectrum Disorder (ASD) classification. The proposed network seamlessly integrates a Multi-Output TSK (MO-TSK) fuzzy system, a Stacked Autoencoder (SAE), and a Transformer. This integration aims to reduce the impact of uncertainty factors, enhance the model's interpretability, and leverage the multi-head attention mechanisms to capture both local and global information, thus strengthening feature learning. Additionally, we explore the utilization of a Contrastive Learning (CL) strategy to further optimize the model. By incorporating these advancements, our AT-STTSK network offers a promising approach for improving the accuracy and reliability of ASD diagnosis based on rs-fMRI data. The validity of the method in this paper is verified on ABIDE dataset, and the result shows that the accuracy rate can reach 85.7% at most.
ISSN:2156-1133
DOI:10.1109/BIBM62325.2024.10822467