Dissertation / PhD Thesis

FZJ-2023-03406

http://join2-wiki.gsi.de/foswiki/pub/Main/Artwork/join2_logo100x88.png Brain Region-wise Connectivity-based Psychometric Prediction: Framework, Interpretation, Replicability and Generalizability

Wu, J.FZJ*

2023

Abstract: The study of brain-behavior relationships is a fundamental aspect of neuroscience. Recently, it has become increasingly popular to investigate brain-behavior relationships by relating the interindividual variability in psychometric measure to the interindividual variability in brain imaging data. In particular, prediction approaches with cross-validation can be useful for identifying generalizable brain-behavior relationships in a data-driven manner. Nevertheless, it remains to be ascertained what brain-behavior relationships can be interpreted from the prediction models, and how generalizable the models are to fully new cohorts. In this work, we attempt to fill in the gap of interpretability by developing a region-wise \acrfull{cbpp} framework. This framework involves a region-wise approach where a prediction model is estimated and evaluated for each brain region. The prediction accuracy of each region-wise model is a direct indication of that brain region's association with the psychometric measure predicted. In study 1, we applied the framework to a range of psychometric variables from a large healthy cohort and demonstrated the helpfulness of the framework in constructing region-wise psychometric prediction profiles or psychometric-wise prediction pattern across the brain. In study 2, we demonstrated the usefulness of the framework in assessing cross-cohort replicability and generalizability in terms of brain-behavior relationships derived from the prediction models, instead of just based on prediction accuracies. In study 3, we systematically examined existing psychometric prediction studies, summarizing the trends in the field, calling for the use of large cohorts and external validation. Overall, our work suggested the importance of interpretability and generalizability for psychometric prediction, recommending the use of multiple large cohorts in evaluating the interpretability and generalizability.

Note: Dissertation, Heinrich-Heine-Universität Düsseldorf, 2023

---

Contributing Institute(s):

1. Gehirn & Verhalten (INM-7)

Research Program(s):

1. 5253 - Neuroimaging (POF4-525) (POF4-525)
2. 5251 - Multilevel Brain Organization and Variability (POF4-525) (POF4-525)

Appears in the scientific report 2023

---