@misc{item_3307369,
  title = {{Improving 3D convolutional neural network comprehensibility via interactive visualization of relevance maps: Evaluation in Alzheimer\textquotesingles disease}},
  abstract = {{Although convolutional neural networks (CNN) achieve high diagnostic accuracy for detecting Alzheimer\textquotesingles 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. 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\textequals663 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 N\textequals1655 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. Across three independent datasets, group separation showed high accuracy for AD dementia vs. controls (AUC$\geq$0.92) and moderate accuracy for MCI vs. controls (AUC$\approx$0.75). Relevance maps indicated that hippocampal atrophy was considered as the most informative factor for AD detection, with additional contributions from atrophy in other cortical and subcortical regions. Relevance scores within the hippocampus were highly correlated with hippocampal volumes (Pearson\textquotesingles r$\approx$-0.81). 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 and high performance achieved in independent samples support the validity of the CNN models in the detection of AD-related MRI abnormalities.}},
  year = {2021},
  slug = {item_3307369},
  author = {Dyrba, M and Hanzig, M and Altenstein, S and Bader, S and Ballarini, T and Brosseron, F and Buerger, K and Cantr\'e, D and Dobisch, L and D\"uzel, E and Ewers, M and Fliessbach, K and Glanz, W and Haynes, JD and Heneka, MT and Janowitz, D and Keles, DB and Kilimann, I and Laske, C and Maier, F and Metzger, CD and Munk, MH and Perneczky, R and Peters, O and Preis, L and Priller, J and Rauchmann, B and Roy, N and Scheffler, K and Schneider, A and Schott, BH and Spottke, A and Spruth, EJ and Weber, M-A and Ertl-Wagner, B and Wagner, M and Wiltfang, J and Jessen, F and Teipel, SJ}
}