Resolving autism spectrum disorder (ASD) through brain topologies using fMRI dataset with multi-layer perceptron...

Area of Science:

• Neuroscience
• Computer Science
• Medical Imaging

Background:

• Autism Spectrum Disorder (ASD) is a neurodevelopmental condition impacting social interaction and communication.
• Previous prediction models for ASD using brain network data have shown limited accuracy.
• Early identification is crucial for timely intervention and support.

Purpose of the Study:

• To develop and evaluate a computer-aided algorithm for early identification of ASD.
• To differentiate between individuals with ASD and typically developing (TD) individuals using fMRI data.
• To identify key brain network features indicative of ASD.

Main Methods:

• Utilized a Multi-Layer Perceptron (MLP) model incorporating logistic regression on fMRI-derived connectivity matrices.
• Employed an AND operation to select statistically significant features across multiple logistic regression analyses.
• Assessed feature importance and model performance on various data subsets to mitigate overfitting.

Main Results:

• The MLP model achieved an accuracy of 83.57% and an AUC of 0.978.
• Identified specific correlations in the Left/Right Lateral Occipital Cortex (in ASD) and Left/Right Cerebellum Tonsil (in TD) as key discriminators.
• MLP classifier demonstrated a recall of 82.61%, outperforming logistic regression (72.46% accuracy).

Conclusions:

• The proposed computer-aided approach, particularly the MLP model, shows significant promise for accurate early detection of ASD.
• Feature analysis revealed distinct brain network patterns associated with ASD and TD individuals.
• This methodology offers a more dependable estimation of feature importance for ASD classification.