Multi-model order spatially constrained ICA reveals highly replicable group differences and consistent predictive
Xing Meng1, Armin Iraji1, Zening Fu1
1Tri-Institutional Center for Translational Research in Neuroimaging and Data Science (TReNDS), Georgia State, Georgia Tech, Emory University, Atlanta, GA, USA.
Neuroimage. Clinical
|May 20, 2023
Summary
This study introduces a new multi-scale analysis for brain functional networks from fMRI data in schizophrenia. The method successfully identified consistent differences in brain connectivity, aiding in biomarker discovery.
Area of Science:
- Neuroimaging
- Computational Neuroscience
- Psychiatric Disorders
Background:
- Resting-state fMRI reveals brain functional networks with potential for biomarker discovery in disorders like schizophrenia.
- Replication challenges in schizophrenia studies stem from disorder complexity, short scan times, and limited data mining capabilities.
- Existing methods like ICA and atlas-based approaches have limitations in comparability and sensitivity to individual variability.
Purpose of the Study:
- To develop and validate a novel multi-objective optimization approach for extracting subject-specific intrinsic connectivity networks (ICNs) at multiple spatial scales from fMRI data.
- To enable the study of interactions across different spatial scales of functional brain networks.
- To assess the utility of this multi-scale framework for identifying biomarkers in schizophrenia.
Main Methods:
- Developed multi-objective optimization scICA with reference algorithm (MOO-ICAR) for multi-scale ICN extraction.
- Applied the framework to a large cohort (N > 1,600) of schizophrenia patients and controls, using separate validation and replication sets.
- Analyzed multiscale functional network connectivity (msFNC) for group differences, classification, and correlation with positive symptoms.
Main Results:
- Highly consistent group differences in msFNC were observed in the cerebellum, thalamus, and motor/auditory networks across datasets.
- Multiple msFNC pairs linking different spatial scales were identified as significant.
- A classification model achieved high performance (85% F1 score, 83% precision, 88% recall) in distinguishing schizophrenia patients from controls.
Conclusions:
- The proposed MOO-ICAR framework demonstrates robustness in evaluating multi-scale brain functional connectivity in schizophrenia.
- The study identified consistent and replicable brain network alterations associated with schizophrenia.
- This approach shows promise for leveraging resting-state fMRI for developing brain biomarkers for psychiatric disorders.
