Presentation (CTI)

OBJECTIVES: Adolescence represents a critical neurodevelopmental window during which the clinical phenotypes of Autism Spectrum Disorder (ASD) and Attention Deficit and Hyperactivity Disorder (ADHD) exhibit substantial symptom overlap, particularly in domains of executive function, social cognition, and impulse control. The differential diagnosis between these two neurodevelopmental conditions in adolescent populations poses considerable clinical challenges, as conventional behavioral assessments often fail to capture the nuanced neurophysiological distinctions. We aim to develop a lightweight deep neural network, Ado2DNet, to accurately discriminate ASD from ADHD in adolescent individuals.
METHODS: Our methodology benchmarked Ado2DNet against established pre-trained architectures, including MobileNet, Xception, and ResNet18, for classification on resting-state fMRI images. The Ado2DNet architecture was systematically designed with optimized convolutional layers, strategically placed maxpooling operations, and adaptive dropout regularization. The network architecture incorporated specialized considerations for adolescent resting-state fMRI data characteristics, including the developmental trajectory of functional connectivity patterns and age-related neural maturation markers.
RESULTS: Experimental results demonstrated that Ado2DNet achieved superior classification performance on the testing set, attaining a classification accuracy of 0.8333, a precision of 0.8163, a recall of 0.8602, and an F1-score of 0.8377. Comparative analysis revealed that the proposed model significantly outperformed conventional pre-trained deep neural network architectures, establishing its superior discriminative capability in distinguishing adolescent ASD from ADHD.
CONCLUSIONS: This system demonstrates significant clinical potential for differential diagnosis between ASD and ADHD in adolescent populations. It enhances the accuracy of clinical decision-making processes while facilitating timely therapeutic interventions that may improve patients' long-term developmental outcomes. Furthermore, its computationally efficient design enables multi-platform deployment, supporting real-time analysis of patient data to expedite diagnostic evaluations and optimize healthcare delivery efficiency for adolescent neurodevelopmental disorders.