% 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”.

@ARTICLE{Dyrba:163520,
      author       = {Dyrba, Martin and Hanzig, Moritz and Altenstein, Slawek and
                      Bader, Sebastian and Ballarini, Tommaso and Brosseron,
                      Frederic and Bürger, Katharina and Cantré, Daniel and
                      Dechent, Peter and Dobisch, Laura and Düzel, Emrah and
                      Ewers, Michael and Fliessbach, Klaus and Glanz, Wenzel and
                      Haynes, John-Dylan and Heneka, Michael T and Janowitz,
                      Daniel and Keles, Deniz B and Kilimann, Ingo and Laske,
                      Christoph and Maier, Franziska and Metzger, Coraline D and
                      Munk, Matthias and Perneczky, Robert and Peters, Oliver and
                      Preis, Lukas and Priller, Josef and Rauchmann, Boris and
                      Roy, Nina and Scheffler, Klaus and Schneider, Anja and
                      Schott, Björn H and Spottke, Annika and Spruth, Eike J and
                      Weber, Marc-André and Ertl-Wagner, Birgit and Wagner,
                      Michael and Wiltfang, Jens and Jessen, Frank and Teipel,
                      Stefan J},
      title        = {{I}mproving 3{D} convolutional neural network
                      comprehensibility via interactive visualization of relevance
                      maps: evaluation in {A}lzheimer's disease.},
      journal      = {Alzheimer's research $\&$ therapy},
      volume       = {13},
      number       = {1},
      issn         = {1758-9193},
      address      = {London},
      publisher    = {BioMed Central},
      reportid     = {DZNE-2022-00280},
      pages        = {191},
      year         = {2021},
      abstract     = {Although convolutional neural networks (CNNs) achieve high
                      diagnostic accuracy for detecting Alzheimer's disease (AD)
                      dementia based on magnetic resonance imaging (MRI) scans,
                      they are not yet applied in clinical routine. One important
                      reason for this is a lack of model comprehensibility.
                      Recently developed visualization methods for deriving CNN
                      relevance maps may help to fill this gap as they allow the
                      visualization of key input image features that drive the
                      decision of the model. We investigated whether models with
                      higher accuracy also rely more on discriminative brain
                      regions predefined by prior knowledge.We trained a CNN for
                      the detection of AD in N = 663 T1-weighted MRI scans of
                      patients with dementia and amnestic mild cognitive
                      impairment (MCI) and verified the accuracy of the models via
                      cross-validation and in three independent samples including
                      in total N = 1655 cases. We evaluated the association of
                      relevance scores and hippocampus volume to validate the
                      clinical utility of this approach. To improve model
                      comprehensibility, we implemented an interactive
                      visualization of 3D CNN relevance maps, thereby allowing
                      intuitive model inspection.Across the three independent
                      datasets, group separation showed high accuracy for AD
                      dementia versus controls (AUC ≥ 0.91) and moderate
                      accuracy for amnestic MCI versus controls (AUC ≈ 0.74).
                      Relevance maps indicated that hippocampal atrophy was
                      considered the most informative factor for AD detection,
                      with additional contributions from atrophy in other cortical
                      and subcortical regions. Relevance scores within the
                      hippocampus were highly correlated with hippocampal volumes
                      (Pearson's r ≈ -0.86, p < 0.001).The relevance maps
                      highlighted atrophy in regions that we had hypothesized a
                      priori. This strengthens the comprehensibility of the CNN
                      models, which were trained in a purely data-driven manner
                      based on the scans and diagnosis labels. The high
                      hippocampus relevance scores as well as the high performance
                      achieved in independent samples support the validity of the
                      CNN models in the detection of AD-related MRI abnormalities.
                      The presented data-driven and hypothesis-free CNN modeling
                      approach might provide a useful tool to automatically derive
                      discriminative features for complex diagnostic tasks where
                      clear clinical criteria are still missing, for instance for
                      the differential diagnosis between various types of
                      dementia.},
      keywords     = {Alzheimer Disease: diagnostic imaging / Cognitive
                      Dysfunction: diagnostic imaging / Humans / Magnetic
                      Resonance Imaging: methods / Neural Networks, Computer /
                      Neuroimaging: methods / Alzheimer’s disease (Other) /
                      Convolutional neural network (Other) / Deep learning (Other)
                      / Layer-wise relevance propagation (Other) / MRI (Other)},
      cin          = {AG Teipel / AG Endres / AG Wagner / Biomarker / AG Speck /
                      AG Simons / Patient Studies Bonn / AG Peters / Core ICRU /
                      AG Düzel / Magdeburg common / AG Gasser / AG Wiltfang / AG
                      Jessen / AG Klockgether / Clinical Research (Munich) / AG
                      Priller / AG Dichgans / AG Schneider / Delcode},
      ddc          = {610},
      cid          = {I:(DE-2719)1510100 / I:(DE-2719)1811005 /
                      I:(DE-2719)1011201 / I:(DE-2719)1011301 / I:(DE-2719)1340009
                      / I:(DE-2719)1110008 / I:(DE-2719)1011101 /
                      I:(DE-2719)5000000 / I:(DE-2719)1240005 / I:(DE-2719)5000006
                      / I:(DE-2719)6000015 / I:(DE-2719)1210000 /
                      I:(DE-2719)1410006 / I:(DE-2719)1011102 / I:(DE-2719)1011001
                      / I:(DE-2719)1111015 / I:(DE-2719)5000007 /
                      I:(DE-2719)5000022 / I:(DE-2719)1011305 /
                      I:(DE-2719)5000034},
      pnm          = {353 - Clinical and Health Care Research (POF4-353) / 351 -
                      Brain Function (POF4-351) / 352 - Disease Mechanisms
                      (POF4-352)},
      pid          = {G:(DE-HGF)POF4-353 / G:(DE-HGF)POF4-351 /
                      G:(DE-HGF)POF4-352},
      experiment   = {EXP:(DE-2719)DELCODE-20140101},
      typ          = {PUB:(DE-HGF)16},
      pubmed       = {pmid:34814936},
      pmc          = {pmc:PMC8611898},
      doi          = {10.1186/s13195-021-00924-2},
      url          = {https://pub.dzne.de/record/163520},
}