Q-Space Novelty Detection with Variational Autoencoders

Abstract

In machine learning, novelty detection is the task of identifying novel unseen data. During training, only samples from the normal class are available. Test samples are classified as normal or abnormal by assignment of a novelty score. The usage of deep neural networks for novelty detection remains an open challenge. Here we propose novelty detection methods based on training variational autoencoders (VAEs) on normal data. We apply these methods to magnetic resonance imaging, namely to the detection of diffusion-space (q-space) abnormalities in diffusion MRI scans of multiple sclerosis patients. q-Space novelty detection can reduce scan time duration and does not require any disease-specific prior knowledge, thus overcoming the disadvantages of other diffusion MRI processing methods. The methods proposed herein outperform the state of the art on q-space data in terms of quality and inference time. Our methods also outperform the state of the art on a standard novelty detection benchmark, and hence are also promising for non-MRI novelty detection.