Classication of MRI Migraine Medical Data using 3D Convolutional Neural Network Hwei Geok Ng 1 , Matthias Kerzel 1 , Jan Mehnert 2 , Arne May 2 , and Stefan Wermter 1 1 Universit¨at Hamburg, Department of Informatics, Knowledge Technology, Vogt-K¨olln-Str. 30, 22527 Hamburg, Germany {5ng, kerzel, wermter}@informatik.uni-hamburg.de 2 Institut f¨ ur Systemische Neurowissenschaften, Universit¨atsklinikum Hamburg-Eppendorf, Martinistraße 52, 20246 Hamburg, Germany {j.mehnert, a.may}@uke.de Abstract. While statistical approaches are being implemented in med- ical data analyses because of their high accuracy and eciency, the use of deep learning computations can potentially provide out-of-the-box in- sights, especially when statistical approaches did not yield a good result. In this paper we classify migraine and non-migraine magnetic resonance imaging (MRI) data, using a deep learning method named convolutional neural network (CNN). 198 MRI scans, which were obtained equally from both data groups, resulted in the maximum classication test accuracy of 85% (validation accuracy: ¯ x=0.69, σ=0.06), compared to the base- line statistical accuracy of 50%. We then used class activation mapping (CAM) method to visualize brain regions that the CNN model took to distinguish one data group from the other and the visualization pointed at the parietal lobe, corpus callosum, brain stem and anterior cingulate cortex, of which the brain stem was mentioned in the medical ndings for white matter abnormalities. Our ndings suggest that CNN and CAM combined can be a useful image-based data analysis tool to add inspira- tion or discussion in the medical problem-solving process. Keywords: Convolutional Neural Network · Class Activation Mapping · Migraine · Magnetic Resonance Imaging. 1 Introduction Statistical approaches are used in medical data analyses because they are ecient to be implemented and return precise results. Nevertheless, given sucient mean- ingful data and computational power, deep learning approaches can also assist in the data analytics process. Convolutional neural networks (CNNs) are a useful deep learning approach, known for their high accuracy in learning relevant fea- tures for arbitrary classication tasks, especially for image classication. CNNs The authors 1 gratefully acknowledge partial support from the German Research Foundation DFG under project CML (TRR 169). Citation: H. G. Ng, M. Kerzel, J. Mehnert, A. May, and S. Wermter, “Classication of mri migraine medical data using 3d convolutional neural network,” in International Conference on Articial Neural Networks - ICANN 2018, V. Kůrková, Y. Manolopoulos, B. Hammer, L. Iliadis, and M. Ilias, Eds. Springer, 2018, pp. 300–309.