ASD-DiagNet

ASD-DiagNet classifies Autism Spectrum Disorder from functional Magnetic Resonance Imaging (fMRI) data by extracting features with an autoencoder and classifying them using a single-layer perceptron to improve diagnostic accuracy.

Key Features:

• Machine Learning Framework: Employs a hybrid learning approach integrating an autoencoder with a single-layer perceptron (SLP) for feature extraction and classification.
• Data Augmentation Strategy: Generates synthetic feature vectors through linear interpolation of existing feature vectors to mitigate small fMRI sample sizes.
• Efficiency and Performance: On the Autism Brain Imaging Data Exchange (ABIDE) public dataset (1,035 subjects from 17 centers) it reported up to a 28% increase in classification accuracy reaching 82% and reduced execution time from approximately seven hours to about 40 minutes.

Scientific Applications:

• Biomarker identification: Identification of quantitative biomarkers associated with ASD from fMRI-derived features.
• Diagnostic classification: Assist classification between ASD subjects and healthy controls using fMRI features for quantitative diagnostic indicators.
• Neurobiological research and early intervention: Provide insights into neurobiological underpinnings of autism and support improved screening that may facilitate early intervention strategies.

Methodology:

Feature extraction using an autoencoder; classification using a single-layer perceptron; data augmentation via linear interpolation of feature vectors.