The mathematics of clinical diagnosis: cognitively-inspired computational psychiatry
Medicine has been advanced by diagnostic technologies such as artificial neural networks that can diagnose cancer from radiological images. However, these diagnostic technologies have mainly been applied to the diagnosis of physical illness, even though mental illness is as relevant as a diagnostic problem in our everyday lives as physical illness. Diagnostic technologies have not been applied to the diagnosis of mental illness for good reason. Psychological diagnosis does not depend on an analysis of physical symptoms. Rather, psychological diagnosis depends on interpreting and understanding the language people use to describe their thoughts and emotions. However, language is a complex and imprecise presentation of mental health. To solve these problems, I evaluated established models of distributed semantics and machine learning classification models to build a computational system that can diagnose people’s mental health from their written language. The system was trained and tested on database of essays written by 1016 participants who also completed five standard measures of mental health. The work joins a growing effort to translate basic cognitive psychology and computational psychology research into the design of cognitive technologies capable of solving complex real-world problems.
categorization, clinical diagnosis, LSA, machine learning, computational psychiatry