An Adaptive Transfer Learning Framework for Multimodal Autism Spectrum Disorder Diagnosis

Area of Science:

• Neuroscience
• Genetics
• Medical Imaging

Background:

• Autism Spectrum Disorder (ASD) is a complex neurodevelopmental condition with diverse characteristics.
• Early ASD diagnosis is challenging due to its heterogeneity and limitations of unimodal approaches.

Purpose of the Study:

• To introduce an adaptive multimodal fusion framework for integrating behavioral, genetic, and structural MRI (sMRI) data for improved ASD diagnosis.
• To overcome the limitations of unimodal diagnostic methods by capturing cross-modal dependencies.

Main Methods:

• Behavioral data analysis using ensemble classifiers with stacking and attention mechanisms.
• Genetic data analysis via Gradient Boosting.
• Structural MRI (sMRI) analysis employing a Hybrid Convolutional Neural Network-Graph Neural Network (Hybrid-CNN-GNN).
• Adaptive late fusion strategy using a Multilayer Perceptron (MLP) to integrate modalities.

Main Results:

• Individual modality accuracies: Behavioral (95.5%), Genetic (86.6%), sMRI (96.32%).
• The integrated multimodal framework achieved a superior diagnostic accuracy of 98.7%.
• The proposed framework demonstrated strong generalization across heterogeneous datasets.

Conclusions:

• The adaptive multimodal fusion framework offers a promising approach for reliable ASD diagnosis.
• This integrated model effectively addresses the limitations of unimodal diagnostic strategies.
• The study highlights the potential of combining diverse data types for enhanced neurodevelopmental disorder diagnosis.