Thesis: Statistical Models on Manifolds for Anomaly Detection in Medical Images

PhD defense: Florian TILQUIN

Title: Statistical Models on Manifolds for Anomaly Detection in Medical Images

Team: IMAGeS

Abstract: Anomaly detection  is a broad task of computer vision and image analysis, with applications ranging from crowd video analysis , to medical image analysis for pathology detection. In our context, we refer to anomaly detection as part of “one versus N” testing: learning a model of normality over a control group, and checking whether a new sample is normal or not (and if so, where). Further on, we will mathematically associate this problem with the one of finding the projection function over the topological space in which our control group lies using non-linear machine learning algorithms, and is the main objective of this thesis. Our topic of application is medical image analysis, with the detection of pathological areas in an image by comparison to a control group, therefore both allowing a classification of the subject as pathological or healthy, and highlighting how each imaged area is afflicted by the pathology. We focus on Alzheimer’s disease (AD), the main neurodegenerative dementia, which affects spatially located areas of the brains such as the hippocampuses, an organ involved in the patient’s memory. However,  our modelization could be applied to any kind of anomaly detection task.

The jury comprises of:

The defense will be held in English on wednesday November the 13th 2019 at 2:00 PM at the "amphithéâtre 2" of the Illkirch IUT.