Reformulation of Clinical Depression Symptoms Using Semantically Enhanced Multilingual Lexical Network رسالة ماجستير

Abstract

Clinical Depression screening is one of the biggest challenges facing psychiatrists, due to the long costly screening process and mental illness-related stigma that creates a serious barrier to mental healthcare. As per World Health Organization (WHO), clinical depression is widely spread and impacts over 264 million individuals around the world and has a large effect on the annual global economy as the lost productivity is estimated at $1 trillion. Therefore, the utilization of approved tested scales in screening, diagnosis, and monitoring response to treatment is important to provide effective needed care. Among the available self-assessment instruments, the Beck depression inventory-II (BDI-II) is one of the most extensively used screening questionnaires for clinical depression. BDI-II, which has 21 categories, shows reliability in discriminating between depressed and non-depressed subjects and identifying the depression severity. The current BDI-II questions’ categories were defined in 1996 and have been updated by the DSM-V diagnostic criteria for diagnosing mental health disorders. BDI-II categories which have not been changed since then, are not statistically well divided into independent clusters as they are based on semantically subjective and correlated measures. These non-structured techniques have relatively low accuracy. To help decrease the treatment gap, it's becoming important to use technology to develop a depression screening application based on previously identified questions by psychiatrists. In this context, natural language processing (NLP) techniques can provide an efficient screening tool to predict clinical depression symptoms and their severity among diagnosed patients. In this research project, we aim to utilize NLP techniques for constructing a lexical-semantic network that enriches the BDI-II questionnaire through the exploitation of extrinsic resources, including WordNet, social media, and unified IV medical language system (UMLS). The enrichment is represented by a reformulation of the original questions and their order using semantically derived relationships from the exploited extrinsic resources. Accordingly, we developed a bilingual screening tool using Arabic and English languages to prove more accurate screening and severity diagnosis to help provide the required mental and physical care and lead to efficacious therapy. The tool is designed to be used by patients who can score the questionnaires themselves without seeing a doctor, while the output network offers an immediate utility that can be used by any other researchers who are studying clinical depression symptoms and their impact in a patient-centric manner. In comparison with the current question categories, the developed tool aims to make the screening process faster and more reliable as it keeps only semantic types that are appropriate for clinical depression. It also aims to increase the accuracy of the screening questionnaire by re-grouping the questions that express the same medical semantics based on multiple clustering models