Related Experiment Video
Updated: Aug 4, 2025
07:12
Using Informational Connectivity to Measure the Synchronous Emergence of fMRI Multi-voxel Information Across Time
Published on: July 1, 2014
12.4K
Autism spectrum disorder recognition based on multi-view ensemble learning with multi-site fMRI.
Li Kang1,2, Jin Chen1,2, Jianjun Huang1,2
1College of Electronics and Information Engineering, Shenzhen University, Shenzhen, 518061 China.
Cognitive Neurodynamics
|April 3, 2023
Summary
This study introduces a novel deep learning network to accurately identify Autism Spectrum Disorder (ASD) using multi-site functional MRI data. The multi-view ensemble approach effectively overcomes data heterogeneity, improving diagnostic performance for ASD.
Area of Science:
- Neuroscience
- Machine Learning
- Biomedical Imaging
Background:
- Autism Spectrum Disorder (ASD) is a neurodevelopmental condition characterized by social and behavioral challenges.
- Early diagnosis of ASD is crucial for effective intervention and improved outcomes.
- Multi-site neuroimaging studies, like those using functional MRI (fMRI), offer larger sample sizes but face challenges with inter-site data heterogeneity.
Purpose of the Study:
- To develop and validate a deep learning-based multi-view ensemble network for improved classification of ASD from normal controls (NC) using multi-site fMRI data.
- To address the performance degradation caused by inter-site heterogeneity in large-scale fMRI datasets.
- To enhance the identification of ASD by integrating diverse functional brain features.
Main Methods:
- A novel multi-view ensemble learning network integrating deep learning was proposed.
- The LSTM-Conv model was employed to extract dynamic spatiotemporal features from fMRI time series.
- Principal Component Analysis (PCA) and a denoising autoencoder were used to extract low/high-level functional connectivity features.
- Feature selection and ensemble learning were applied to combine these multi-view features.
Main Results:
- The proposed method achieved a 72% classification accuracy for ASD vs. NC on multi-site ABIDE dataset fMRI data.
- Leave-one-out cross-validation on single-site data demonstrated strong generalization capability.
- The highest classification accuracy reached 92.9% on the CMU site data, highlighting robust performance.
Conclusions:
- The multi-view ensemble learning network effectively improves ASD classification performance by leveraging diverse fMRI data features.
- This approach mitigates issues arising from data heterogeneity in multi-site neuroimaging studies.
- The method shows significant potential for accurate and reliable ASD identification in clinical settings.
