Related Experiment Video
Updated: Feb 17, 2026
14:27
Identification of Disease-related Spatial Covariance Patterns using Neuroimaging Data
Published on: June 26, 2013
16.3K
Multisite Machine Learning Analysis Provides a Robust Structural Imaging Signature of Schizophrenia Detectable Across
Martin Rozycki1, Theodore D Satterthwaite1,2, Nikolaos Koutsouleris3
1Center for Biomedical Image Computing and Analytics, Department of Radiology, University of Pennsylvania Perelman School of Medicine, Philadelphia, PA.
Schizophrenia Bulletin
|November 30, 2017
Summary
Structural brain imaging reveals a reliable signature for schizophrenia, enabling accurate individual diagnosis. This neuroimaging biomarker shows promise for clinical application and correlates with symptom severity.
Area of Science:
- Neuroimaging
- Psychiatry
- Machine Learning
Background:
- Previous studies indicate widespread structural brain abnormalities in schizophrenia using regional volumetry and machine learning.
- Clinical utility of structural imaging biomarkers requires integration of high-dimensional data and reproducible results across diverse populations and individuals.
Purpose of the Study:
- To establish a robust and reproducible neuroanatomical signature for schizophrenia using pooled multi-site data.
- To validate the signature's generalizability across different sites, populations, and scanners for single-patient classification.
Main Methods:
- Utilized advanced multivariate analysis tools on pooled case-control imaging data from 5 sites (941 participants, 440 with schizophrenia).
- Performed analyses at regional volume, voxelwise, and distributed pattern scales.
- Tested single-subject classification with single-site, pooled-site, and leave-site-out cross-validation.
Main Results:
- Identified a widespread pattern of reduced gray matter volume (medial prefrontal, temporolimbic, peri-Sylvian cortex) and enlarged ventricles/pallidum.
- Achieved 76% cross-validated prediction accuracy (AUC=0.84) using pooled data.
- Leave-site-out validation demonstrated robust generalizability (accuracy/AUC 72-77%/0.73-0.91).
- Individual classifications correlated significantly with negative symptoms.
Conclusions:
- Structural neuroimaging data can provide a robust and reproducible imaging signature for schizophrenia.
- The developed signature shows strong generalizability and potential for clinical application.
- A web portal is available for community access to individualized magnetic resonance imaging classifications.
