Re-Categorizing the Attention Deficit Hyperactivity Disorder SNAP-IV Diagnostic Tool into Concisely Semantic Groups by the use of Natural Language Processing رسالة ماجستير

Abstract

The diagnostic for the attention deficit hyperactivity (ADHD) disorder based on a symptoms rating questionnaire conducted using interviews is clinically recommended as a diagnostic measure, the SNAP-IV rating scale is a version from Swanson, Nolan and Pelham (SNAP) to measure the core symptoms of ADHD. where it requires answering 90 questions, there is significant semantic overlap across questions, the current questions categories are based on semantically subjective and correlated measures rather than well-defined mathematically independent clusters, the accuracy of non-structured approaches is very limited. The purpose of this study is to improve the utility of SNAP-IV by deriving the hidden semantic dimensions that are encoded in the original questions by understanding the semantic basis based on medical knowledge resources to: i) reduce the length by decreasing questions, ii) Increase the accuracy by grouping questions that express the same medical concept, iii) Enabling healthcare professionals to decide retaining or removing overlapped questions. We conducted several clustering models. Four different prototypes built based on the produced categories from each one of them and been used in a real world completed and scored dataset for ADHD diagnostic. We ran statistical independent-sample t-tests to compare the averages between children with ADHD and matched controls. Results show that all new produced categories from each clustering approach shows significant difference. The semantically-enhanced medical enrichment terms which is based on a medical knowledge graph has proved to produce a new concise concept based categorization with less number of questions without harming the clinical procedure.