% 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},
}