2025, Vol. 5, Issue 2, Part A

Autism Spectrum Disorder (ASD) is a neurodevelopmental disorder that presents with a range of social, communicative, and behavioral challenges. The heterogeneity of the disorder makes its diagnosis complex, relying primarily on behavioral assessments that are subjective and prone to inconsistencies. Early diagnosis is critical for effective intervention, and the need for objective biomarkers to assist in this process has led to increasing interest in neuroimaging studies. Structural magnetic resonance imaging (sMRI), functional MRI (fMRI), and Diffusion Tensor Imaging (DTI) have emerged as promising tools for identifying neurobiological markers that could aid in the diagnosis of ASD. However, despite significant advancements in neuroimaging research, identifying consistent and reliable biomarkers remains a challenge due to the diversity in symptomatology and brain structure-function relationships in ASD.

In recent years, machine learning (ML) and deep learning (DL) techniques, particularly Convolutional Neural Networks (CNNs), have shown exceptional promise in the domain of medical image analysis. CNNs, known for their ability to automatically learn hierarchical features from raw image data, are particularly suited for neuroimaging applications where patterns in brain structure and function may not be readily apparent through traditional statistical methods. This study investigates the application of CNNs to neuroimaging data—specifically structural MRI, functional MRI, and DTI—to identify novel diagnostic biomarkers for ASD.

The dataset used in this study consists of neuroimaging data from 200 subjects, including 100 individuals diagnosed with ASD and 100 neurotypical controls. The study applies a multi-modal approach, combining data from structural, functional, and diffusion tensor imaging to provide a comprehensive view of the brain’s anatomy, connectivity, and activity. The results of the CNN-based analysis indicate that ASD can be classified with high accuracy (93%) when combining the three imaging modalities. Structural MRI contributed the most to the classification accuracy, followed by functional MRI and DTI.

This paper discusses the potential of CNN-based analysis to uncover novel biomarkers for ASD and highlights its superiority over traditional machine learning techniques. Furthermore, the findings underscore the importance of integrating multiple neuroimaging modalities to enhance diagnostic accuracy. The study paves the way for future research to refine deep learning models for clinical applications, suggesting the potential for CNN-based neuroimaging biomarkers to aid in the early, objective diagnosis of ASD, leading to better-tailored interventions.