% IMPORTANT: The following is UTF-8 encoded.  This means that in the presence
% of non-ASCII characters, it will not work with BibTeX 0.99 or older.
% Instead, you should use an up-to-date BibTeX implementation like “bibtex8” or
% “biber”.

@PHDTHESIS{Komeyer:1050728,
      author       = {Komeyer, Vera},
      title        = {{N}eurobiological {P}redictors of {H}and {G}rip {S}trength
                      as a {G}lobal {H}ealth {M}arker: {M}ethodological
                      {F}oundations and {I}nterpretable {B}rain-{B}ehaviour
                      {P}rediction in {L}arge-{S}cale {N}euroimaging},
      school       = {HHU Düsseldorf},
      type         = {Dissertation},
      reportid     = {FZJ-2026-00471},
      pages        = {151},
      year         = {2026},
      note         = {Dissertation, HHU Düsseldorf, 2026},
      abstract     = {Aging populations worldwide are widening the gap between
                      lifespan and healthspan, underscoring theneed for early,
                      scalable markers of organismal health and intervention
                      targets before overt clinicaldecline. Hand grip strength
                      (HGS) has emerged as a highly reliable and low-cost
                      predictor of systemwidefactors such as frailty, cognitive
                      decline, and mortality. Despite its simplicity and
                      clearmusculoskeletal determinants, explaining its
                      system-wide predictive value requires a deeperunderstanding
                      of underlying brain-level neurobiological architectures,
                      which is currently lacking. Thisthesis addresses this gap by
                      investigating generalizable neural predictors of HGS using
                      machine learning(ML) with large-scale, multi-modal
                      neuroimaging data, grounded in methodological foundations
                      forinterpretable brain-behaviour prediction.A critical
                      review of methodological constraints and potential
                      mitigation strategies inobservational neuroimaging-based ML
                      studies was conducted to promote more reliable
                      andgeneralizable brain-behaviour predictions and
                      interpretations (Study 1). Such studies can be hamperedby
                      pitfalls including data leakage, site-effects in multi-site
                      datasets, misleading post-hoc modelinterpretations arising
                      from feature multicollinearity, and model bias due to
                      confounding. Strict out-ofsampleevaluation and
                      clustering-based interpretation to deal with feature
                      multicollinearity wereidentified as suitable solutions. To
                      support principled confounder selection, a theoretically
                      informed butempirically pragmatic 3-step approach was
                      developped (Study 2). The proposed approach
                      integratesmethdology from causal inference - domain
                      knowledge, directed acyclic graphs, and respective
                      graphrules - with associative data-driven modeling.Building
                      on these foundations, a comprehensive, interpretable
                      multi-modal predictive workflowrevealed generalizable,
                      system-level neuroimaging predictors of HGS in a large,
                      healthy cohort fromthe UK Biobank (Study 3). Across modeling
                      approaches, microstructural integrity – particularly
                      inascending medial lemniscus, thalamic radiations, and
                      associative white-matter pathways – as well assubcortical
                      gray matter volume (GMV), mainly in the anterior globus
                      pallidus, emerged as relevantcontributors. In contrast,
                      cortical structural measures and functional imaging features
                      contributed littleto predictive performance. Collectively,
                      these findings position HGS as a behavioural readout of
                      thebrain’s capacity to coordinate, and integrate
                      information across motor, sensory, cognitive,
                      andmotivational systems, rather than a purely peripheral
                      muscle measure or an isolated motor output.In sum, this
                      thesis establishes a framework for neurobiologically
                      interpretable large-scale brainbehaviourprediction and
                      applies it to elucidate why HGS functions as a powerful
                      marker of globalhealth. By intergating methodological rigor
                      with system-level neuroimaging, it demonstrates howsimple
                      behavioural phenotypes can serve as informative windows into
                      the functioning and integrity ofdistributed neural
                      architectures. Future work should determine whether
                      identified HGS-linked neuralsignatures provide better or
                      earlier prognostic and interventional value than the
                      behavioural measureitself.},
      cin          = {INM-7},
      cid          = {I:(DE-Juel1)INM-7-20090406},
      pnm          = {5253 - Neuroimaging (POF4-525)},
      pid          = {G:(DE-HGF)POF4-5253},
      typ          = {PUB:(DE-HGF)11},
      doi          = {10.34734/FZJ-2026-00471},
      url          = {https://juser.fz-juelich.de/record/1050728},
}