% 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”.

@INPROCEEDINGS{Huysegoms:1048785,
      author       = {Huysegoms, Marcel and Kropp, Jan-Oliver and Wenzel, Susanne
                      and Oliveira, Sarah and Paquola, Casey and Amunts, Katrin
                      and Dickscheid, Timo},
      title        = {{T}owards a {C}ellular 3{D} {M}odel of the {H}uman {B}rain
                      at 1µm {I}sotropic {R}esolution},
      reportid     = {FZJ-2025-04900},
      year         = {2025},
      abstract     = {<b>Background:</b> 3D histological reconstruction is well
                      established at mesoscopic scales (~20µm). At this
                      resolution, the task can be framed as an optimization of
                      spatial transformations driven by pixel-level similarity
                      metrics, typically within coarse-to-fine reconstruction
                      frameworks that integrate both slice-to-slice and
                      histology-to-reference registrations (Amunts et al., 2013;
                      Alkemade et al., 2022).<br><Pushing toward cellular
                      precision (<5µm) introduces several key challenges. First,
                      intensity-based matching becomes unreliable: individual dark
                      features often represent different cells across consecutive
                      sections, so intensities do not correspond one-to-one.
                      Moreover, staining variability and tissue distortions
                      further amplify noise. Second, reference volumes (e.g. MRI
                      or blockface) lack the necessary spatial detail to constrain
                      alignment at the microstructural level. In the absence of a
                      reliable reference volume, sequential reconstruction alone
                      is prone to cumulative alignment drift (commonly known as
                      the “banana effect”), which can artificially straighten
                      curved anatomical structures.<br>We aimed to address these
                      challenges through the development of a novel framework for
                      3D reconstruction of histological sections that accurately
                      represents the original volume using a three-step
                      feature-matching and global optimization
                      strategy.<br><br><b>Methods:</b> Optical z-scanning enables
                      imaging of twenty depths within each 20 µm thick
                      histological section by gradually shifting the focal plane,
                      resulting in a 1 µm isotropic resolution volume per
                      section. The principal challenge then lies in accurately
                      reconstructing the sections to achieve a faithful
                      representation of the anatomy of the unsliced brain. Our
                      approach involves (i) computing point correspondences
                      between neighbouring sections (e.g. bisected cells or blood
                      vessels). Then, (ii) optimize a global cost function to
                      jointly estimate the spatial coordinates of matched features
                      across all sections, enforcing both matching and smoothness
                      constraints. Finally, (iii) to produce the final volume,
                      polyaffine transformations are fitted to warp each
                      histological section onto the optimized feature
                      positions.<br><br><b>Results:</b> We compared our global
                      reconstruction approach against several other strategies,
                      demonstrating its potential to outperform standard
                      sequential pipelines in terms of alignment accuracy,
                      structural smoothness, and anatomical preservation. The
                      final reconstructed volume maintains natural cortical
                      curvature, resolves laminar organization throughout the
                      cortex (even in oblique cutting planes) and preserves
                      overall brain shape.<br><br><b>Conclusion:</b> Our framework
                      provides a robust, scalable foundation for building
                      cellular-resolution 3D models of the human brain. It makes
                      minimal assumptions about feature types, accommodating both
                      traditional descriptors (e.g., SURF) and ML-based detectors
                      (e.g., LightGlue). The method also generalizes across
                      different tissue stainings and section thicknesses (10–60
                      µm). Although the approach can operate without a reference
                      volume, one can be incorporated to better preserve overall
                      shape in cases of substantial nonlinear tissue deformation.
                      Future work will generate detailed, region-specific
                      descriptions of cellular organization, which can in turn
                      support realistic simulations of human brain activity.},
      month         = {Oct},
      date          = {2025-10-27},
      organization  = {9th BigBrain Workshop - HIBALL Closing
                       Symposium, Berlin (Germany), 27 Oct
                       2025 - 29 Oct 2025},
      subtyp        = {After Call},
      cin          = {INM-1},
      cid          = {I:(DE-Juel1)INM-1-20090406},
      pnm          = {5254 - Neuroscientific Data Analytics and AI (POF4-525) /
                      5251 - Multilevel Brain Organization and Variability
                      (POF4-525) / HIBALL - Helmholtz International BigBrain
                      Analytics and Learning Laboratory (HIBALL) (InterLabs-0015)
                      / EBRAINS 2.0 - EBRAINS 2.0: A Research Infrastructure to
                      Advance Neuroscience and Brain Health (101147319) /
                      Helmholtz AI - Helmholtz Artificial Intelligence
                      Coordination Unit – Local Unit FZJ (E.40401.62)},
      pid          = {G:(DE-HGF)POF4-5254 / G:(DE-HGF)POF4-5251 /
                      G:(DE-HGF)InterLabs-0015 / G:(EU-Grant)101147319 /
                      G:(DE-Juel-1)E.40401.62},
      typ          = {PUB:(DE-HGF)24},
      url          = {https://juser.fz-juelich.de/record/1048785},
}