Entropy-based analysis is increasingly used in task-based functional magnetic resonance imaging (fMRI) to quantify neural signal complexity and information dynamics, but variation in entropy definitions, parameter choices, and analytic scope can limit cross-study comparability. To systematically review how entropy measures are implemented, parameterized, and interpreted in task-based fMRI studies in healthy human subjects, focusing on methodological practice. Web of Science was searched using the keywords “fMRI” and “entropy” for the period 2000–2023, restricted to journal articles, proceedings papers, review articles, meeting abstracts, and book chapters. Included studies used task-based fMRI, applied entropy-based quantitative measures, involved healthy human participants, and reported original empirical findings or methodological applications. Non-human, clinical, and resting-state studies were excluded. Records were screened by verifying whether “fMRI” and “entropy” appeared in the title, keywords, Keywords Plus, or abstract. Extracted items included entropy type, analytic scope (regional/voxel-wise, network-level, connectivity-based), parameter and reporting details, task types, and preprocessing context where available. Data were synthesized using structured narrative methods because meta-analysis was not appropriate given differences in entropy definitions, parameterization, task types, and outcome metrics. Risk of bias was assessed with an adapted Joanna Briggs Institute (JBI) checklist (Joanna Briggs Institute, 2017). Database searches yielded 1,313 records. 274 were screened and 234 full texts assessed. 92 studies met inclusion criteria. Exclusions at full-text were primarily resting-state studies (n = 81), clinical populations (n = 42), and non-human studies (n = 19). Across the 92 included studies, Shannon entropy predominated (78.3%), followed by sample entropy (9.78%), transfer entropy (4.35%), multiscale entropy (3.26%), approximate entropy (3.26%), and multiple-entropy approaches (1.09%). Entropy measures were found to be matched with distinct methodological roles. Approximate and sample entropy were commonly used for regional or voxel-wise signal regularity, multiscale entropy for multi–time scale complexity (often at the network level), transfer entropy for directed connectivity, and Shannon entropy for broad applications including machine-learning feature and validation use. Evidence synthesis was constrained by inconsistency in entropy formulations, parameter reporting, preprocessing decisions, and outcome metrics. Formal heterogeneity testing, subgroup analyses, and sensitivity analyses were not conducted, and results were summarized descriptively. Task-based fMRI entropy research is methodologically diverse but consistently demonstrates the feasibility of using entropy to characterize task-related brain complexity across different analytic levels. The prevalent use of Shannon entropy and inconsistent parameter/reporting practices underscore the need for clearer, standardized reporting and reproducible implementation guidance to improve comparability across studies.
Park et al. (Thu,) studied this question.