ASD-GraphNet: A novel graph learning approach for Autism Spectrum Disorder diagnosis using fMRI data.

Researchers created a computer program called ASD-GraphNet that looks at brain scans (fMRI) to help diagnose autism. The program maps how different brain areas connect and communicate with each other. When tested, it correctly identified autism in about 75% of cases. This could potentially help doctors make more objective autism diagnoses using brain scans alongside the usual behavioral observations.

This study introduces ASD-GraphNet, a machine learning framework that analyzes brain connectivity patterns from fMRI scans to diagnose Autism Spectrum Disorder. Using data from the ABIDE database, researchers developed a graph-based approach that maps brain networks using established brain atlases. The system extracts multiple types of features from these brain networks and applies various machine learning classifiers to distinguish between individuals with ASD and neurotypical controls. The framework achieved 75.25% accuracy in classification, suggesting potential for objective, brain-based diagnostic tools that could complement traditional behavioral assessments for ASD diagnosis.

ASD-GraphNet represents a promising step toward objective neuroimaging-based diagnostic tools for ASD. The 75% accuracy suggests potential clinical utility but requires validation in larger, diverse samples and comparison with standard diagnostic methods before clinical implementation.

Sample size not reported. Single dataset used (ABIDE). No comparison to clinical diagnostic accuracy. Limited discussion of generalizability across populations or scanner types. Performance metrics beyond accuracy not detailed.

Autism Spectrum Disorder (ASD) is a complex neurodevelopmental condition with heterogeneous symptomatology, making accurate diagnosis challenging. Traditional methods rely on subjective behavioral assessments, often overlooking subtle neural biomarkers. This study introduces ASD-GraphNet, a novel graph-based learning framework for diagnosing ASD using functional Magnetic Resonance Imaging (fMRI) data. Leveraging the Autism Brain Imaging Data Exchange (ABIDE) dataset, ASD-GraphNet constructs brain networks based on established atlases (Craddock 200, AAL, and Dosenbach 160) to capture intricate connectivity patterns.

The framework employs systematic preprocessing, graph construction, and advanced feature extraction to derive node-level, edge-level, and graph-level metrics. Feature engineering techniques, including Mutual Information-based selection and Principal Component Analysis (PCA), are applied to enhance classification performance. ASD-GraphNet evaluates a range of classifiers, including Logistic Regression, Support Vector Machines, and ensemble methods like XGBoost and LightGBM, achieving an accuracy of 75.25% in distinguishing individuals with ASD from healthy controls. This demonstrates the framework's potential to provide objective, data-driven diagnostics based solely on resting-state fMRI data.

By integrating graph-based learning with neuroimaging and addressing dataset imbalance, ASD-GraphNet offers a scalable and interpretable solution for early ASD detection, paving the way for more reliable interventions. The GitHub repository for this project is available at: https://github.com/AmirDavoodi/ASD-GraphNet.