Towards the Interpretability of Deep Learning Models for Human Neuroimaging

The scientific work “Towards the Interpretability of Deep Learning Models for Human Neuroimaging“, with the support of iToBoS project, has been published.

“Towards the Interpretability of Deep Learning Models for Human Neuroimaging”.

  • Authors: Simon M. Hofmann *,1,2,3, Frauke Beyer 1,3, Sebastian Lapuschkin 2 , Markus Loeffler 4 , KlausRobert Müller 5,6,7,8,9, Arno Villringer 1,3,10,11, Wojciech Samek 2,9, A. Veronica Witte *,1,3.
  • Affiliations :1 Department of Neurology, Max Planck Institute for Human Cognitive and Brain Sciences, 04103 Leipzig, Germany 2 Department of Artificial Intelligence, Fraunhofer Institute Heinrich Hertz, 10587 Berlin, Germany 3 Clinic for Cognitive Neurology, University of Leipzig Medical Center, 04103 Leipzig, Germany 4 IMISE, University of Leipzig, 04103 Leipzig, Germany 5 Machine Learning Group, Technical University Berlin, 10623 Berlin, Germany 6 Department of Artificial Intelligence, Korea University, 02841 Seoul, South Korea 7 Brain Team, Google Research, 10117 Berlin, Germany 8 Max Planck Institute for Informatics, 66123 Saarbrücken, Germany 9 BIFOLD - Berlin Institute for the Foundations of Learning and Data, 10587 Berlin, Germany 10 MindBrainBody Institute, Berlin School of Mind and Brain, Humboldt-Universität zu Berlin, 10099 Berlin, Germany 11 Center for Stroke Research, Charité – Universitätsmedizin Berlin, 10117 Berlin, Germany.


Brain-age (BA) estimates based on deep learning are increasingly used as neuroimaging biomarker for brain health; however, the underlying neural features have remained unclear. We combined ensembles of convolutional neural networks with Layer-wise Relevance Propagation (LRP) to detect which brain features contribute to BA. Trained on magnetic resonance imaging (MRI) data of a population-based study (n=2637, 18-82 years), our models estimated age accurately based on single and multiple modalities, regionally restricted and whole-brain images (mean absolute errors 3.37-3.86 years). We find that BA estimates capture aging at both small and large-scale changes, revealing gross enlargements of ventricles and subarachnoid spaces, as well as lesions, iron accumulations and atrophies that appear throughout the brain. Divergence from expected aging reflected cardiovascular risk factors and accelerated aging was more pronounced in the frontal lobe. Applying LRP, our study demonstrates how superior deep learning models detect brain-aging in healthy and at-risk individuals throughout adulthood.


This work is supported by the European Union, European Regional Development Fund as part of the LIFE-LIFT project, and the Free State of Saxony within the framework of the excellence initiative, and LIFE-Leipzig Research Center for Civilization Diseases, University of Leipzig (project numbers 713-241202, 14505/2470), and by the German Research Foundation (project numbers 209933838 CRC1052 Obesity mechanisms A1 and WI 3342/3-1). Further support was provided by the German Ministry for Education and Research (BMBF) through Berlin Institute for the Foundations of Learning and Data (BIFOLD; refs. 01IS18025A and 01IS18037A), MALT III (ref. 01IS17058), Patho234 (ref. 031L0207D) and Transparent Medical Expert Companion (TraMeExCo, ref. 01IS18056A), European Union's Horizon 2020 research and innovation programme through Intelligent Total Body Scanner for Early Detection of Melanoma (iToBoS, grant agreement No 965221), as well as the Grants 01GQ1115 and 01GQ0850; and by Deutsche Forschungsgemeinschaft (DFG) under Grant Math+, EXC 2046/1, Project ID 390685689; by the Institute of Information & Communications Technology Planning & Evaluation (IITP) grant funded by the Korea Government (No. 2019-0-00079, Artificial Intelligence Graduate School Program, Korea University).

Get the complete article at biorxiv website.