A CCA+ICA based model for multi-task brain imaging data fusion and its application to schizophrenia

Area of Science:

• Neuroimaging
• Biomarker Discovery
• Multivariate Statistics

Background:

• Multi-task brain imaging data collection is increasingly common in medical research.
• Existing methods for fusing multi-task data may lack accuracy or clear connections between datasets.

Purpose of the Study:

• To propose and validate a novel multi-task data fusion model, CCA+ICA, for enhanced analysis of brain imaging data.
• To assess the model's performance in identifying group-discriminative brain networks and correlations with clinical factors.

Main Methods:

• Developed a joint blind source separation (BSS) model, CCA+ICA, integrating canonical correlation analysis (CCA) and independent component analysis (ICA).
• Applied the model to simulated and real functional magnetic resonance imaging (fMRI) data from schizophrenia patients and healthy controls performing sensorimotor (SM) and Sternberg working memory (SB) tasks.
• Compared CCA+ICA with other joint BSS models and analyzed correlations with duration of illness.

Main Results:

• CCA+ICA demonstrated high estimation accuracy and correctly connected datasets with common or distinct correlations.
• The model revealed a significant negative correlation between duration of illness and temporal lobe activation in schizophrenia patients.
• CCA+ICA identified sensorimotor cortex as group-discriminative regions for both tasks and specific networks (superior temporal gyrus in SM, prefrontal cortex in SB) as task-specific.

Conclusions:

• CCA+ICA is an effective tool for multi-task data fusion, offering improved biomarker identification from brain imaging.
• The model provides consistent results and fills a gap in existing multivariate methods for complex neuroimaging analyses.
• Findings highlight the utility of CCA+ICA in uncovering task-specific and group-discriminative brain networks and their clinical relevance.