Recognition of autism in subcortical brain volumetric images using autoencoding-based region selection method and...

Area of Science:

• Neuroimaging and Machine Learning
• Developmental Neuroscience

Background:

• Autism Spectrum Disorder (ASD) diagnosis remains challenging due to symptom variability and subjective assessment.
• Advanced neuroimaging and machine learning offer potential for improved ASD detection.
• Focusing on subcortical brain regions is crucial for understanding ASD pathology.

Purpose of the Study:

• To enhance Autism Spectrum Disorder (ASD) recognition using brain MRI data.
• To improve the accuracy and interpretability of ASD diagnosis through advanced computational methods.
• To identify critical subcortical brain regions associated with ASD.

Main Methods:

• Subcortical structures were extracted from brain MRI datasets.
• A 3D autoencoder was utilized to identify ASD-related brain regions.
• Feature selection methods and a Siamese Convolutional Neural Network (SCNN) were employed for classification.

Main Results:

• The 3D autoencoder successfully identified and reconstructed significant subcortical regions in ASD.
• Regions such as the Putamen and Pallidum showed high agreement, indicating their importance in distinguishing ASD.
• The SCNN classifier, using Mutual Information features, achieved a classification accuracy of 0.66.

Conclusions:

• A two-stage model combining autoencoder and SCNN significantly improves ASD classification from brain MRI.
• Iterative feature extraction enhances the identification of ASD-related brain areas.
• The proposed approach improves classification performance and neuroimaging data interpretability.