Identification of multimodal brain imaging association via a parameter decomposition based sparse multi-view...

Area of Science:

• Neuroimaging
• Biostatistics
• Computational Neuroscience

Background:

• Multimodal brain imaging data offers complementary insights but contains intertwined shared and specific information.
• Distinguishing shared from specific information is crucial for comprehensive brain disease characterization.
• Existing methods often fall short in effectively dissecting this complex information.

Purpose of the Study:

• To propose a novel method, parameter decomposition based sparse multi-view canonical correlation analysis (PDSMCCA), for analyzing multimodal brain imaging data.
• To effectively identify both modality-shared and modality-specific information within these datasets.
• To enhance the understanding of complex brain disease pathologies.

Main Methods:

• Developed a parameter decomposition based sparse multi-view canonical correlation analysis (PDSMCCA) approach.
• Applied PDSMCCA to synthetic and real neuroimaging data.
• Evaluated the method's ability to perform feature selection and identify multi-view associations.

Main Results:

• PDSMCCA achieved higher correlation coefficients compared to the standard multi-view canonical correlation analysis (SMCCA) method.
• The method demonstrated superior canonical weights on both synthetic and real neuroimaging datasets.
• PDSMCCA showed effective modality-shared and -specific feature selection capabilities, improving multi-view association identification.

Conclusions:

• Parameter decomposition is confirmed as a viable strategy for identifying modality-shared and -specific imaging features.
• The PDSMCCA method enhances the analysis of multimodal associations.
• Leveraging diverse information from multimodal imaging data aids in better understanding brain diseases like Alzheimer's disease.