% 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{Boztoprak:1033600,
      author       = {Boztoprak, Zeynep and Amunts, Katrin and Dickscheid, Timo
                      and Schiffer, Christian},
      title        = {{T}owards {D}ecoding {F}iber {A}rchitecture in the {H}uman
                      {B}rain {U}sing {D}eep {L}earning},
      reportid     = {FZJ-2024-06480},
      year         = {2024},
      abstract     = {Analyzing the structural organization of the human brain
                      involves the study of nerve fibers that connect neurons or
                      whole brain regions. Characterizing the arrangement of these
                      anatomical connections, also known as fiber architecture,
                      contributes to a better understanding of how the brain
                      processes information. Three-dimensional Polarized Light
                      Imaging (3D-PLI) allows the visualization of these short-
                      and long-range single nerve fibers and fiber bundles,
                      providing detailed images of fiber organization at the
                      microscopic level. However, analyzing and interpreting the
                      large amounts of complex data is a time-consuming task that
                      requires expert knowledge and a lot of computational power,
                      making manual engineering difficult and leading us to use a
                      data-driven approach to automatically learn descriptive
                      features of nerve fibers.<br><br>Previous work has
                      demonstrated the ability of deep learning methods in
                      identifying characteristics of nerve fibers in the vervet
                      monkey brain, such as boundaries between gray and white
                      matter and specific U-fibers. Building upon these promising
                      results, our present study extends this research to focus on
                      the adult human brain.<br><br>We train a deep neural network
                      to map high-resolution image patches extracted from 3D-PLI
                      sections to feature vectors that encode the fiber
                      architectonic properties encoded in the patch. Exploiting
                      the fact that spatially close locations in the brain tend to
                      share architectonic properties, we train the neural network
                      using a contrastive learning objective that promotes the
                      mapping of spatially close image patches to similar feature
                      representations. The model is trained on 30.000 patches
                      (size 2048x2048 px at 2.66 px/m) from 26 3D-PLI sections.
                      Learned features are then analyzed by visualizing their
                      corresponding U-MAP embeddings in 2D.<br><br>Our analysis
                      shows that features from patches sampled from different
                      sections tend to cluster based on section number rather than
                      architectural similarity. In particular, features of patches
                      from distant sections are also distant in feature space,
                      suggesting that the network captures section-specific
                      features rather than the underlying brain architecture. We
                      hypothesize that this effect is due to the uneven
                      distribution of the available PLI-scanned sections in
                      certain parts of the brain (non-isotropic), which affects
                      the spatial consistency of the data. As we do not observe
                      clustering that reflects anatomical features, suggesting
                      that the current model fails to capture these structural
                      properties, future work will address this challenge by
                      exploring data augmentation techniques to introduce
                      variability and counteract differences between sections and
                      sampling strategies. These approaches aim to encourage the
                      network to focus on architectural features of the brain
                      rather than section-specific features.},
      month         = {Nov},
      date          = {2024-11-19},
      organization  = {INM Retreat 2024, Jülich (Germany),
                       19 Nov 2024 - 20 Nov 2024},
      subtyp        = {After Call},
      cin          = {INM-1},
      cid          = {I:(DE-Juel1)INM-1-20090406},
      pnm          = {5251 - Multilevel Brain Organization and Variability
                      (POF4-525) / 5254 - Neuroscientific Data Analytics and AI
                      (POF4-525) / X-BRAIN (ZT-I-PF-4-061) / HIBALL - Helmholtz
                      International BigBrain Analytics and Learning Laboratory
                      (HIBALL) (InterLabs-0015) / Helmholtz AI - Helmholtz
                      Artificial Intelligence Coordination Unit – Local Unit FZJ
                      (E.40401.62)},
      pid          = {G:(DE-HGF)POF4-5251 / G:(DE-HGF)POF4-5254 /
                      G:(DE-HGF)ZT-I-PF-4-061 / G:(DE-HGF)InterLabs-0015 /
                      G:(DE-Juel-1)E.40401.62},
      typ          = {PUB:(DE-HGF)24},
      url          = {https://juser.fz-juelich.de/record/1033600},
}