Information Fusion for Stroke Clinical Decision Support: Architectures, Algorithms, and Applications

Stroke clinical decision support systems (CDSSs) increasingly rely on the integration of heterogeneous data sources, including neuroimaging, clinical records, physiological signals, and contextual information. Effectively fusing such multimodal, asynchronous, and partially incomplete data remains a central challenge, directly influencing diagnostic accuracy, treatment selection, and outcome prediction. While a growing body of literature has explored machine learning and deep learning solutions for stroke CDSSs, existing studies are often fragmented with respect to information fusion strategies and architectural design choices. This review provides a comprehensive and fusion-centric synthesis of multimodal stroke CDSS research, focusing on how information fusion is performed across different stages of the modeling pipeline. We systematically categorize fusion approaches into early (data-level), feature-level (joint representation), late (decision-level), and hybrid paradigms, and analyze their relationships with major clinical tasks such as diagnosis, triage, reperfusion decision support, outcome prediction, and rehabilitation forecasting. Particular attention is given to modality-specific representation learning, architectural modularity, temporal and multi-task fusion, and emerging uncertainty-aware and graph-based methods. By emphasizing architectural abstraction rather than application-specific implementations, this work bridges methodological advances in information fusion with practical clinical decision support requirements. We further identify key limitations, open challenges, and future research directions necessary for developing robust, interpretable, and clinically deployable multimodal fusion systems for stroke care.

کلیدواژه ها:

Information Fusion ، Stroke clinical decision support ، Multimodal Learning ، Representation learning ، Medical decision systems