Machine Learning Techniques for Supporting Decision Making in Differential Diagnosis

The general objective is to complete a pilot study into the feasibility of using computational algorithms of natural language processing (NPL) to analyze the free speech of psychiatric patients in order to detect language and prosody-based variables. The first step involves gathering data, the second entails selecting proper NLP techniques, and the third deploying software tools to instantiate metrics and models. The data will be obtained from de-identified information from the Indiana psychiatric illness interview, guided clinical interviews, structured clinical interviews, dynamic interviews, and thematic apperception tests.
As Latent Semantic Analysis has aided in predicting the risk of future psychotic episodes in an at risk population (Bedi et al., 2015), we hypothesize that other techniques of NLP will help to identify linguistic features for various psychiatric states found in mood disorders, personality disorders, and the spectrum of psychosis. We will explore the extent to which these computational tools are able to function simultaneously on the level of content – semantic incoherence, mentalization, fear of understanding, doing and undoing-, as well as at the level of form: literality, derailment, thought blocking, perseveration, and underlying grammatical structure.
In our times, technology, data gathering and decision support models are more and more integrated into society, with artists such as Gibson and Vinge foretelling of a society where technological advances rework social links, forms of communication, and increase body plasticity. In the rapidly expanding field of Computer Science on the approach towards AIs, we consider the integration of psychoanalytic knowledge with machine learning algorithms to be quite important.