Multimodal Machine Learning for Enhanced Autism Spectrum Disorder Detection

Abstract

Autism Spectrum Disorder (ASD), a complex neurodevelopmental condition, poses a significant diagnostic challenge due to its heterogeneous clinical presentation. Traditional diagnostic methods often rely on subjective behavioral assessments, which can be time-consuming and prone to human error. To address these limitations, this thesis presents a novel framework for the enhanced and objective detection of ASD using Multimodal Machine Learning (MML). Our approach integrates multiple data modalities—including facial expressions, vocal patterns, and eye-gaze tracking data—to capture a more holistic and nuanced representation of ASD-related behaviors. We employ deep learning architectures, such as Convolutional Neural Networks (CNNs) for image data and Recurrent Neural Networks (RNNs) for sequential audio data, fused through an innovative attention-based fusion mechanism. This mechanism dynamically weights the importance of each modality, improving the model`s robustness and diagnostic accuracy. The proposed model is trained and validated on a diverse dataset of pediatric subjects, achieving a superior diagnostic accuracy of over 95%, outperforming unimodal and traditional machine learning approaches. Our findings demonstrate that the synergy of multimodal data significantly enhances the diagnostic precision and offers a more reliable, scalable, and non-invasive tool for early ASD screening. This research contributes to the development of a powerful, data-driven diagnostic aid that can support clinicians and facilitate earlier intervention, ultimately improving the quality of life for individuals with ASD.