PANDAS: Paediatric Attention-Deficit Hyperactivity/Disorder Application Software.
Thesis (MEng)--Stellenbosch University, 2018.
The published article for this Doctoral/Master’s is available at [http://hdl.handle.net/10019.1/123072]
ENGLISH ABSTRACT: Attention-deficit/hyperactivity disorder (ADHD) is a common neuropsychiatric disorder that impairs social, academic, and occupational functioning in children, adolescents and adults. It is estimated that approximately as high as 10% of South African children have ADHD. Some dilemmas are however present in terms of the treatment of the disorder: firstly, there are no risk-free methods for its treatment and secondly, no fully objective diagnostic assessments exist. To date, very few quantitative methods have been successfully implemented. It is therefore necessary to further investigate methods that objectively diagnose, treat, and manage the disorder. The aim of the study is thus to develop a novel method that can be used as an aid to provide screening of ADHD. The method proposed is the form of a tablet-based game with underlying algorithms. The objective of the method is to differentiate between an ADHD individual versus a non-ADHD individual, based on the way they play the game. A beta-testing phase was done and comprised of 30 children (19 non-ADHD and 11 ADHD) between the ages of 4 and 18 years old. The machine learning model that was used was linear support vector machine (SVM). Two datasets were used: 1) game-play dataset which included data such as task completion time and number of mistakes made and 2) accelerometer data set from the tri-axial accelerometer. A feature set was extracted from these two datasets and the best features were selected using sequential forward selection. These best features were then used for developing the classifier. A test set accuracy of 85.7% was achieved. Leave-one-out cross-validation (LOOCV) was performed and its accuracy was 83.5%. An overall classification accuracy of 86.5% was achieved. For the application of a screening tool, sensitivity was deemed an important metric and. The model achieved a sensitivity of 75% which was seen as acceptable. The results of the classifier were indicative that a quantitative tool could indeed be developed to screen for ADHD.
AFRIKAANSE OPSOMMING: Aandagafleibaarheid-hiperaktiwiteitsindroom (ADHD) is 'n algemene neuropsigiatriese versteuring wat sosiale-, akademiese- en beroepsfunksionering belemmer by jong kinders, tieners en volwassenes. Dit is beraam dat ongeveer as meer as 10% van Suid-Afrikaanse kinders ADHD het. Sommige dilemmas is egter teenwoordig met betrekking tot die behandeling van die versteuring: eerstens is daar geen risiko-vrye metodes van behandeling nie en tweedens bestaan daar geen volledige objektiewe diagnostiese assessering nie. Tot vandag toe is daar baie min kwantitatiewe metodes wat suksesvol geimplenteer is. Daarom is dit dus nodig om verdere metodes te ondersoek wat die versteuring objektief kan diagnoseer, behandel en bestuur. Die mikpunt van die studie is dus om 'n nuwe metode te ontwikkel wat gebruik kan word as 'n hulpmiddel om vinnige en akkurate keuring van ADHD te voorsien. Die voorgestelde metode is in die vorm van 'n elektroniese tablet-gebasseerde speletjie met onderliggende algoritmes. Die doel van die metode is om te onderskei tussen 'n individu met ADHD en 'n individu sonder ADHD, gebasseer op die manier hoe hulle die speletjie speel. 'n Beta-toets fase wat bestaan uit 30 kinders tussen die ouderdomme van 4 en 18 jaar oud was gedoen (19 met ADHD en 11 sonder ADHD). Die masjienleer model wat gebruik vir die studie was support vector machine (SVM). 'n Test akkuraatheid van 85.7% was behaal. Leaveone- out cross-validation was gebruik word en 'n akkuraatheid van 83.5% was behaal. Ter afsluiting was dit gedemonstreer dat 'n kwantitatiewe hulpmiddel ontwikkel kan word vir die keuring van ADHD.