Using natural language processing in clinical research: The promise and challenges of translating into practical tools

There is much excitement about potential applications of machine learning (ML) and natural language processing (NLP) to detect clinically significant changes in cognitive and mental state. However, the path from lab to clinic is thorny, and a clear map is needed so as to minimize the risks associated with using complex artificial intelligence (AI) models. The four speakers of this symposium - with diverse expertise in neuropsychology, clinical practice, cognitive science, artificial intelligence (AI), natural language processing, and technology transfer - will evaluate what it really will take for the algorithms and models to be translated into practical tools within psychology and psychiatry. The first speaker (Diaz-Asper) will draw on her experience of leveraging NLP methods to detect early cognitive decline in the elderly at risk for Alzheimer’s disease based on only brief samples of speech captured in real world settings. The second speakers (Cohen & Rodriguez) will focus on the need for a completely new approach to psychometrics that both enables the field to move beyond modeling just one data channel to integrate multimodal features and also explicitly build models that are culturally sensitive. The third speaker (Chandler) will challenge current practice of using aggregated data without first establishing ergodicity, namely that group level characteristics generalize to each individual, and the talk will showcase this approach using interactive data visualization techniques. The fourth speaker (Foltz) will consider how these methods and technology in general can best be translated to benefit the clinical world.