Leveraging unsupervised machine learning to map neural mechanisms of psychopathology: a review and user guide

Past and present neuroimaging studies of psychopathology have heavily relied upon general linear models to infer relationships between brain activity and symptoms. However, this work has resulted in little tangible benefit for clinical implementation. In the era of transdiagnostic approaches such as the Research Domain Criteria framework, a subfield of research has emerged which seeks to utilize unsupervised machine learning to map the relationships between neural activity and psychopathology agnostic of existing diagnostic categories. The ability of machine learning algorithms to extract patterns from large amounts of data may hold promise for deriving the underlying patterns that cut across psychiatric disorders. In the present review, we survey neuroimaging studies of psychopathology that have utilized unsupervised learning algorithms. We synthesize findings regarding the structure of psychopathology from results reported by these studies and compare these findings to established taxonomies. Considerations regarding study design and analytical methods are explored, including discussion of key assumptions made in current models. Future directions, as well as suggestions for researchers interested in these methods, are also discussed.

Name thesis:degree_name

M.S.

Level thesis:degree_level

Thesis

Discipline thesis:degree_discipline

Psychology

Grantor

University of Illinois at Urbana-Champaign

Year dc:date

2022

Author dc:creator

• Richier, Corey J.

Contributors dc:contributor

• Heller, Wendy
• Koyejo, Sanmi

• Machine learning
• psychopathology
• neuroimaging

Statement dc:rights

• Copyright 2021 Corey Richier

Language dc:language

en

Handle dc:identifier

http://hdl.handle.net/2142/113211

OAI identifier oai:identifier

oai:www.ideals.illinois.edu:2142/113211