An Autism Spectrum Disorder Identification Method Based on 3D-CNN and Segmented Temporal Decision Network.

Area of Science:

• Neuroscience
• Artificial Intelligence
• Medical Imaging

Background:

• Autism Spectrum Disorder (ASD) presents with social communication deficits and repetitive behaviors.
• Functional MRI (fMRI) is used to study brain abnormalities in ASD, but faces challenges in capturing spatiotemporal dynamics.
• Existing methods struggle with complex fMRI data characteristics and limited temporal information.

Purpose of the Study:

• To develop an advanced framework for improved spatiotemporal dynamics capture in brain activity.
• To enhance the accuracy and efficiency of Autism Spectrum Disorder (ASD) recognition using fMRI data.
• To provide a novel computational tool for ASD diagnosis and understanding neurobiological mechanisms.

Main Methods:

• A novel ASD recognition method combining 3D Convolutional Neural Networks (3D-CNNs) and segmented Long Short-Term Memory (LSTM) networks.
• Automatic extraction of high-dimensional spatial features from 4D fMRI data using 3D-CNNs.
• Classification of concatenated spatiotemporal features using Gradient Boosting Decision Trees (GBDTs) with a voting mechanism.

Main Results:

• The proposed method achieved an average accuracy of 0.85 on the ABIDE dataset.
• Outperformed existing state-of-the-art approaches in Autism Spectrum Disorder (ASD) classification.
• Demonstrated enhanced efficiency in spatiotemporal feature extraction and learning complex brain activity patterns.

Conclusions:

• The developed method offers a new solution for ASD classification by effectively utilizing spatiotemporal information from 4D fMRI data.
• Achieved significant improvements in classification performance, providing a valuable computational tool for ASD diagnosis.
• Offers insights into the neurobiological mechanisms underlying Autism Spectrum Disorder (ASD).