Exploring early stage psychosis through multimodal approaches: a longitudinal study
Date
2024
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
University of Delaware
Abstract
Psychotic disorders are marked by positive and negative symptoms along with functional impairment. Besides clinically-measured features, neuroimaging (via MRI) provides measurements of the brain structure, electroencephalogram (EEG) can measure deficits in the Auditory Steady-State Response (ASSR), and fMRI or EEG can be used to measure the resting state network activity. These measurements from norms across the population (cross-sectionally) or longitudinal changes within an individual are key features that may lead to more targeted intervention (targeted clinical and cognitive remediation therapies) by tracking the individual’s progression trajectory. At early stage psychosis, which is within the first six years of illness, we posit that impairments and changes are already present, but difficult to isolate due to the heterogeneity of the disease and the population. We investigate the relationship among these measurements using statistical modeling and machine learning. We utilize the McLean Hospital First Episode of Psychosis longitudinal dataset, which includes schizophrenia spectrum disorder and affective psychosis patients from ages 17-30, along with matched healthy controls. These subjects undergo a battery of clinical assessments, structural MRI, EEG, and resting state fMRI at baseline and several follow-up timepoints. ☐ In our first work, we decompose structural MRI cortical thickness into easier to interpret features using non-negative matrix factorization (NMF). We elucidate that cortical thinning is already present at early stage psychosis and features are gyral, not functional network based. We further cluster patients based on NMF features to uncover the patterns of thinning and clinical deficits over one year to determine functional outcome. In our second work, we find that the ASSR EEG paradigm demonstrates patient deficits in the 40 Hz at early stage psychosis. Machine learning models are used to separate cases and controls, and identify the channels driving the classification. The top predictors of patient channels’ responses are negative symptoms, community functioning, and manic symptoms. We explore baseline channels’ correlations to short-term functional outcome. Our third work analyzes the longitudinal resting state functional fMRI data using structural equation modelling (SEM), as a proof-of-concept study. The SEM is able to identify regions of interest at both the atlas (brain region) level and functional network level whose functional connectivity changes over time, and can be used to stratify subjects based on this directionality. ☐ The final work is an initial investigation into multimodal strategies to jointly decompose structural MRI, EEG, resting state fMRI, and clinical measures into components that are highly-correlated across modalities. With these holistic components as biomarkers, we hope to better classify ESP versus HC from their multimodal measures, and within ESP, to stratify into deficit versus non-deficit prognoses, to better inform clinical treatment.
Description
Keywords
Early stage psychosis, Electroencephalogram, Functional MRI, Longitudinal dataset, Multimodal measures, Structural MRI