A Convolutional Neural Network-Based Connectivity Enhancement Approach for Autism Spectrum Disorder Detection

Area of Science:

• Neuroscience
• Artificial Intelligence
• Medical Imaging

Background:

• Autism spectrum disorder (ASD) diagnosis remains challenging, necessitating advanced detection methods.
• Existing research suggests both under- and over-connectivity deficits in the autistic brain.
• Previous studies have theoretically supported these connectivity alterations.

Purpose of the Study:

• To develop an enhanced deep learning framework for accurate early diagnosis of ASD.
• To leverage theories of brain under- and over-connectivity in the proposed model.
• To improve upon current ASD detection capabilities.

Main Methods:

• A novel framework combining an enhancement approach with deep learning (Convolutional Neural Networks - CNNs).
• Creation of image-like connectivity matrices from brain data.
• Enhancement of connections associated with identified connectivity alterations.

Main Results:

• The proposed framework achieved a high prediction accuracy of up to 96% for ASD.
• Validation was performed using the large-scale Autism Brain Imaging Data Exchange (ABIDE I) dataset.
• The enhancement approach effectively utilized connectivity properties for improved detection.

Conclusions:

• The developed CNN-based framework shows significant promise for accurate and early ASD diagnosis.
• Enhancing connectivity matrices based on known deficits aids in identifying ASD.
• This approach offers a valuable tool for advancing ASD detection research and clinical application.