PANDA: Paediatric attention-deficit/hyperactivity disorder app

Abstract

The development of a novel method that validates and enhances the current subjective diagnostic methods/tools for attention-deficit/hyperactivity disorder (ADHD) is investigated. The proposed method/tool is in the form of a tablet-based game with underlying artificial intelligence, such as machine learning. Two mini-games were developed, each dealing with one ADHD subtype: the inattentive subtype and the hyperactivity subtype. The objective of each mini-game is to differentiate between an ADHD individual (either from the inattentive or hyperactivity subtype) and a non-ADHD individual, based on game-play data. The design of the mini-games was based on analyzing the ADHD criteria defined by the Diagnostic and Statistical Manual of Mental Disorders (DSM-V) and converting those criteria into measurable parameters, where applicable. Those measurable parameters were then implemented in the mini-games and used as a means to gather objective data. Beta-testing was performed with a population of 40 subjects (20 ADHD-inattentive, 20 ADHD-hyperactivity) between the ages of 4 and 17 years old. A clinical study has begun following mini-games optimisation based on feedback obtained during beta-testing. The clinical study comprises a total of 200 subjects between the ages of 4 and 17 years of age. 156 subjects will be used to train and validate the proposed machine learning algorithms, while the remaining 44 will be used to test the classification accuracy of the algorithms. According to a statistical POWER analysis, it was seen that using a sample size of 156, an ADHD population could be differentiated from a non-ADHD population with 16% error. Given this fact, it is speculated that using neural networks and support vector machines, a smaller error can be expected.