Abstract

In this work, we proposed a novel approach for the detection of anomalies in dMRI scans of the human brain. Diffusion-MRI scans are a powerful technique in medical imaging that has images with 3D diffusion vectors for each voxel in the 3D space as an output. Previous works were able to create a rotation-equivariant convolutional neural network for this kind of “6D”-data. Using and extending this approach, we were able to create a rotation-equivariant anomaly detection model that builds upon the idea of SkipGANomaly and that showed promising results in our experiments.

This research project was carried out together with Rafal Kurda under the supervision of Dr. Vladimir Golkov and Prof. Dr. Daniel Cremers at the Computer Vision Group of TUM.