Επανάλυση της μελέτης «Συνδεσιμότητα των εγκεφαλικών δικτύων στην ήπια έως μέτρια κατάθλιψη» (Bezmaternykh et al., 2021) : αναπαραγωγιμότητα των ευρημάτων σχετικά με τη λειτουργική συνδεσιμότητα σε κατάσταση ηρεμίας με χρήση των εργαλείων fMRIPrep και NeuroMark
Year: 2026
Type: article
Abstract: Resting-state functional connectivity (FC) is widely used to characterize network alterations in major depression, yet reported findings are heterogeneous and depend on analytical choices, raising questions of reproducibility. This thesis is a same-data reanalysis of Bezmaternykh et al. (2021), not an independent replication, examining whether their principal cross-sectional (Study 1: 51 patients, 21 controls) and longitudinal (Study 2: 29 patients, treated and untreated) FC findings survive a standardized, prespecified pipeline. Data were preprocessed with fMRIPrep and additional denoising; networks were defined using the fixed NeuroMark 2.2 triple-network representation (default mode, executive control, salience); connectivity was estimated by zero-lag Pearson correlation; and inference combined the Network-Based Statistic, covariate-adjusted domain-level tests, and equivalence testing (TOST). Robustness used matched sensitivity branches: for Study 1, an alternative fMRIPrep version, denoising variants, a lag-shift estimator, and a paper-style data-driven ICA; for Study 2, a finer NeuroMark 1.0 posterior/anterior DMN parcellation. The reported within-DMN and DMN-ECN differences were not reproduced: effects were small and non-significant (d ≈ 0.12 versus a reported d ≈ 0.61), no connection survived multiple-comparison correction, and equivalence excluded effects of the originally reported magnitude. The longitudinal changes likewise did not reproduce and ran directionally opposite; equivalence was inconclusive for the no-treatment within-DMN change but excluded the treatment-related DMN-ECN effects. Across the seven primary comparisons only three agreed in direction (42.9%, indistinguishable from chance), and near-complete overlap (28 of 29 sharing the same baseline voxel data) precludes treating agreements as independent support. The principal FC findings therefore did not reproduce at the reported magnitude or stability. Non-significance was not treated as absence: where equivalence excluded the reported sizes the results challenge those magnitudes, while others remain unresolved given limited precision. Crucially, conclusions shifted with network definition, indicating that reproducibility is contingent on analytical representation.
Source: Νημερτής
Language: English
Subfield: Cognitive Neuroscience
Field: Neuroscience
Domain: Life Sciences
Open Access status: green
OpenAlex ID: W7224687615
