Applying Artificial Intelligence to Big-data Research: Investigation on obsessive compulsive symptoms as a template

Abstract

Artificial Intelligence (AI) has been widely adopted in various fields, assisting people in saving time, making less efforts, and avoiding mistakes on daily basis. In medical practice, AI also takes an important role on diagnosis, treatment, and evaluation of clinical decision making. In this seminar, the speaker, Dr. Chin-Kuo Chang, is going to present his research at South London and Maudsley NHS Foundation Trust (SLaM) and Institute of Psychiatry, Psychology, and Neuroscience (IoPPN), King’s College London by the application of AI using the technology of Natural Language Processing (NLP) on the electronic health records (EHR) for big-data research. It will start with an introduction for the setting of SLaM hospital and characteristics of its primary, secondary, and even tertiary psychiatric services for local population in Southeast London. Based on the Clinical Record Interactive Search (CRIS) system as the source for massive data with external data linkages, a series of analyses on the physical consequences of severe mental illness (SMI) has been published. Combining the expertise of computer science, artificial intelligence (AI), and linguistics, Natural Language Processing (NLP) is one of the methods to deal with the context of a sentence, mimicking the real-world process of learning a language as human being. We adopted the NLP software, general architecture for text engineering (GATE), to process unstructured text data in EHRs and extract information needed for further analysis. With the rules-based approach, rather than machine learning, we accomplished several tasks for clinical symptoms, patients’ context, interventions, and treatment outcomes. As an example, the development of an App for the detection of obsessive compulsive symptoms (OCS) for people with SMI is raised. The speaker is going to talk about the steps of building such an App, with sufficient details about how to apply the rules to give feedback on the design and validation of the App. The results revealed a precision of 64% and a recall of 76% at document level, with some preliminary analysis on the prediction of mortality for people with SMI shown. Suggestions for further development of this App are thus given. Meanwhile, advanced implementation of NLP on academic publication informatics will be discussed as one of the new directions of application of AI in near future.