Blind Source Separation for Unimodal and Multimodal Brain Networks: A Unifying Framework for Subspace Modeling

Area of Science:

• Neuroscience
• Data Science
• Signal Processing

Background:

• Blind Source Separation (BSS) methods have significantly advanced human brain studies over the past decade.
• Applications span various neuroimaging modalities, focusing on extracting latent networks as underlying sources.
• The proliferation of diverse BSS methods, each with unique objective functions and constraints, has led to complexity and potential confusion for practitioners.

Purpose of the Study:

• To organize the diverse landscape of BSS models in neuroimaging.
• To establish a novel unifying view by exposing key features and connections among models.
• To aid practitioners in selecting appropriate BSS models for their specific neuroimaging applications.

Main Methods:

• Systematic organization of diverse BSS models based on key features, objective functions, and underlying structures.
• Identification and exposition of connections among different BSS models based on their properties and subspace structures.
• Review of current BSS applications in neuroimaging, informed by the unifying framework.

Main Results:

• A novel unifying view of BSS models in neuroimaging is established.
• Important connections between various BSS models are unveiled based on their properties.
• A high-level descriptive structure is provided to guide model selection for practitioners.
• Current applications of BSS in neuroimaging are reviewed, highlighting model interconnections.

Conclusions:

• The organized framework and revealed model connections facilitate better understanding and selection of BSS methods.
• The insights gained provide a broader sense of generalization and highlight directions for future BSS model development.
• Emerging multi-dataset multidimensional (MDM) models show promise for studying the healthy brain and disease-related changes.