TY - THES A1 - Dexl, Jakob Frederik T1 - Mapping of natural learning processes for the development of neural network architectures - Implementation and comparison of approaches in the classification of radiological data sets N2 - In order to provide more transparency on convolutional neural networks (CNN) for education and research purposes this work has three main objectives: • Create CNN models based on two different architectures, which classify magnetic resonance images of the brain into normal and abnormal • Investigate these models by applying state of the art visualization techniques. For this purpose, a simple accessible Application Interface (API) for Keras sequential models will be developed. • Describe the quality of the models based on the visualizations and compare their overall classification to the human classification procedure. T2 - Abbildung natürlicher Lernprozesse für die Entwicklung neuronaler Netzarchitekturen - Implementierung und Vergleich von Ansätzen bei der Klassifizierung eines radiologischen Datensatzes KW - CNN KW - visualization KW - machine learning KW - radiology KW - image classification KW - Neuronales Netz KW - Visualisierung KW - Maschinelles Lernen KW - Radiologie Y1 - 2018 U6 - http://nbn-resolving.de/urn/resolver.pl?urn:nbn:de:bvb:860-opus4-521 ER -