COMBINING PHENOTYPIC AND RESTING-STATE FMRI DATA FOR AUTISM CLASSIFICATION WITH RECURRENT NEURAL NETWORKS

Area of Science:

• Neuroimaging
• Machine Learning
• Developmental Neuroscience

Background:

• Autism spectrum disorder (ASD) identification using resting-state functional magnetic resonance imaging (rsfMRI) is challenging due to ASD heterogeneity.
• Recurrent neural networks (RNNs) show promise for rsfMRI classification, but often exclude valuable phenotypic data.
• Integrating diverse data types like rsfMRI and phenotypic features into deep learning models presents technical hurdles.

Purpose of the Study:

• To develop and evaluate novel deep learning methodologies for integrating phenotypic data with rsfMRI for improved ASD classification.
• To address the challenge of combining heterogeneous data sources within a unified deep learning framework.
• To enhance the accuracy and robustness of automated ASD identification.

Main Methods:

• Proposed several deep learning architectures for multimodal data fusion (rsfMRI time-series and phenotypic features).
• Utilized a recurrent neural network (RNN) framework adapted for combined data inputs.
• Employed a cross-validation strategy on the Autism Brain Imaging Data Exchange (ABIDE) dataset.

Main Results:

• The best-performing model achieved a classification accuracy of 70.1% for identifying ASD.
• The integrated deep learning approach demonstrated superior performance compared to prior methods relying solely on rsfMRI.
• The study successfully combined rsfMRI and phenotypic data within a single deep learning model.

Conclusions:

• Integrating phenotypic data with rsfMRI in a deep learning framework significantly improves autism spectrum disorder classification accuracy.
• The developed methodologies offer a promising approach for more accurate and comprehensive ASD identification.
• This work highlights the potential of multimodal deep learning in understanding and diagnosing complex neurodevelopmental conditions like ASD.