The primary objective of this project is to investigate a system architecture and methodologies for deriving explanations from CNN models to reach a self-explanatory, human-comprehensible neural network. We will address the use case of detecting dementia and mild cognitive impairment due to Alzheimer’s disease and frontotemporal lobar degeneration in magnetic resonance imaging (MRI) data.


Machine learning models must be accurate to be able to provide reasonable explanations.
We validate all models in independent datasets to test their performance.


Machine learning models need to focus on key features that are causally related to the target and often known a priori.
Relevance maps help us to assess the level of noise and to detect bias in the training data.


Machine learning models should provide an explanation describing the decision-making process in an intuitive way.
Our goal is to develop a system architecture that can extract relevant information and dynamically synthesize descriptive explanations.

Deep learning InteractiveVis

In 2020, Martin Dyrba developed a convolutional neural network architecture to detect Alzheimer’s disease in MRI scans. The diagnostic performance was validated in three independent cohorts.

From the neural networks, we can derive relevance maps that indicate the brain areas with high contribution to the diagnostic decision. Medial temporal lobe atrophy was shown as most relevant area which matched our expectations, as hippocampus volume is actually the best established neuroimaging marker for Alzheimer’s disease.

Try the app | Further reading

Project head

Dr. Martin Dyrba
+49 381 494-9482

Welcome to our website, here you can inform yourself basically cookie-free.

We would be pleased if you would allow a cookie to be set for analysis purposes in order to optimise our provided information. All data are pseudonymous and are only used by the DZNE. We deliberately avoid third-party cookies. You can deselect this setting at any time here.

Your browser allows the setting of cookies: