Computational Mental Health

Overview

For my postdoctoral research, I measure symptoms of serious mental illness by analyzing their digital phenotypes (video, audio, questionnaires) using machine learning and natural language processing (NLP) techniques. Most of my work involves developing pipelines for data cleaning, ETL processing, and analysis. These pipelines are developed using a combination of Python, R, and Bash, which are then executed locally and on Microsoft Azure platform.

This position is in partnership with Center for Computation & Technology and the Affective Science and Psychopathology Laboratory (ASAP Lab) at Louiana State University (LSU).

Collaborators and Links

ASAP Lab | Affective Science and Psychopathology Laboratory

We are advancing the assessment and treatment of serious mental illness using highly sophisticated and inexpensive "biobehavioral" technologies.

• Alex Cohen, PhD (PI)
• Tovah Cowan, Graduate Student
• Michael D. Masucci, Graduate Student
• Ole Edvard Granrud, Graduate Student

NLP & Psychopathology Research Group

• Gregory P. Strauss, PhD | University of Georgia
• Peter Foltz, PhD | Research Professor | Institute of Cognitive Science, University of Colorado
• Chelsea Chandler, PhD | Institute of Cognitive Science, University of Colorado
• Brita Elvevåg, PhD | University of Tromsø - The Arctic University of Norway
• Terje Bektesevic Holmlund, PhD | Postdoctoral researcher | University of Tromsø

Natural Language Processing and Mental Health

Natural Language Processing (NLP) is a computational approach to processing, analyzing, and quantifying various aspects of language using machine learning . It has become indispensable to modern society. NLP has also been critical for understanding language, and how it relates to cognition, affect and social functions. It is increasingly being used to “digitally phenotype” aspects of serious mental illness ; an endeavor that could reshape diagnosis, assessment and treatment. NLP allows for automation that can enhance traditional assessment by improving the efficiency, validity and accuracy of data collection, for example, by using data collected using mobile recording devices and social media platforms as individuals navigate their daily routines. The use of high-dimensional or “big” data, collected on large and demographically heterogeneous samples can also help address interpretive and practical constraints on traditional assessments. NLP can facilitate systematic and repeated assessment (e.g., hourly, daily, weekly); important because “naturalistic” measurements can be less affected by learning and practice effects than traditional clinical measures. In all, NLP can improve accuracy for objectifying relatively specific aspects of psychopathology.

Machine Learning Models and Mental Health

Facial expression analysis can support clinical decision making for individuals with Serious Mental Illness (SMI). Current methods rely on the creation and support of clinical judgements and are hampered by the labor-intensive task and potential biases of human-coding. Machine Learning (ML) algorithms, however, are more computationally feasible, highly dynamic, and can be fine-tuned on an individualistic scale. We use feature selection and machine learning techniques to (1) identify a sub-set of facial expression variables that predict critical symptoms associated with SMI and (2) build a model that can identify these symptoms from individual videos. Facial expressions of 6 symptoms were examined using automated facial analysis software and coded by trained human raters. Our models were able to select 20-30 important facial expression variables from over 150 features that were significantly correlated to symptoms such as hostility, depression, and excitement. Further, our models demonstrated high predictive power and convergent validity with clinical and functional variables. Combining computerized facial analysis with ML approaches can provide individualized emotion monitoring information for digital health monitoring systems of affective computing information and ameliorate the arduous task of human coding.