"Leveraging Transfer Learning and Deep Neural Networks for Early Detection of Neurodevelopmental Disorders in Children"

Abstract

Early identification of neurodevelopmental disorders (NDDs) such as Autism Spectrum Disorder (ASD), Attention Deficit Hyperactivity Disorder (ADHD), and developmental delays is critical for timely intervention and improved developmental outcomes. Traditional diagnostic methods often rely on behavioral assessments, which are subjective and time-consuming. This research explores the potential of transfer learning and deep neural networks (DNNs) in detecting NDDs at an early stage using medical imaging and clinical data. By employing pre-trained convolutional neural networks (CNNs) such as AlexNet, VGGNet, and ResNet, and fine-tuning them for pediatric neuroimaging datasets, the study demonstrates significant improvements in classification accuracy and model robustness. The findings indicate that transfer learning reduces training time, mitigates overfitting, and enhances generalization even with limited data availability. Furthermore, the integration of multimodal inputs—including MRI images and behavioral metrics—substantially improved diagnostic sensitivity. This approach promises a transformative shift in pediatric healthcare, supporting clinicians with intelligent tools for early, precise, and scalable NDD screening.