Leveraging temporal pupil dynamics and deep learning for interpretable ADHD diagnosis

Abstract

Attention-deficit/hyperactivity disorder (ADHD) is the most common behavioral condition and the second most common chronic illness in children. This neurodevelopmental disorder characterized by persistent patterns of inattention, hyperactivity, and impulsivity signifi cantly impact daily functioning. Traditional diagnostic methods, such as clinical interviews and behavioral observations, provide valuable insights. However, these methods are often subjective and reliant on observer interpretation, leading to variability in diagnosis and scal ability issues. A biomarker based diagnosis would be an advantage as that would be less error prone & time consuming. Recent studies have underscored the significance of various pupil features in distinguishing between typical and ADHD populations. This research aims to improve the diagnosis of neurodevelopmental disorders by using a smartphones, focusing on the psychology of eyes through pupil dynamics analysis. We are developing a user-friendly smartphone application that would use its camera to capture pupil dynamics and help in the diagnosis of ADHD and ASD while participants are performing standardized cognitive tasks. Our method involves collecting and validating datasets, ensuring the psychological reliability of the data by drawing insights from existing research papers. Building machine learning and deep learning models on an already published dataset, we aim to create a model that uses artificial intelligence to classify and predict neurodevelopmental disorders based on analyzed pupil dynamics. We are further doing research on the interpretability of the models which would further reduce the problem of scalability and accuracy on variety of datasets. This approach not only enhances diagnostic accuracy but also explores the psychological intricacies associated with these disorders.