Predictive Modeling of Childhood Neurodevelopmental Disorders

Overview

Early diagnosis of neurodevelopmental disorders is critical as delayed detection negatively impacts prognosis and long-term outcomes for afflicted individuals. Traditional diagnostic methods are inadequate because they either rely on subjective evaluations or expensive and inaccessible technology. This research developed eye fundus images as an AI-powered biomarker for the early diagnosis of Autism Spectrum Disorder (ASD) and Attention-Deficit/Hyperactivity Disorder (ADHD). The proposed system extracts disorder-related patterns and features from input fundus images through a Convolutional Neural Network (CNN) into feature vectors, which are then analyzed by a fully-connected neural network to output a prediction of normal, ASD, or ADHD. The CNN feature extractor is pre-trained using cosine similarity-based contrastive learning to produce consistent feature vectors between images representing the same disorder. Through transfer learning, this refined CNN is then applied to improve model accuracy and robustness against natural variance in retinal features that can hinder feature vector quality. Through extensive experimentation, it is shown that the proposed method outperforms the existing state-of-the-art CNN in accuracy across all data augmentation configurations. I expect that the proposed method contributes to the use of eye fundus images as an AI-powered biomarker for the early detection of neurodevelopmental disorders.

Competition history

• AJAS 2025 Category not listed

Related projects

Closest projects by meaning, across every fair and year in the corpus.

Browse more like this

• All Behavioural and Social Sciences projects
• AJAS Behavioural and Social Sciences projects
• Every AJAS 2025 project