Many diseases are known to be connected to pathological changes in shape of anatomical organs. One prominent example is the neuro-degenerative Alzheimer's disease that we will focus upon during the hackathon.
Shapes can be represented in Riemannian shape spaces, we aim to obtain a classifier system that takes the non-Euclidean structure of shape spaces into account. The statistical shape model based approach Ambellan et al. 2021 (arxiv) will serve as a starting point. Possible extensions include different metrics (e.g. Log-Cholesky), advanced Riemannian mean-variance analysis, robust estimators for centrality (e.g. med-ian), and intrinsic machine learning approaches (e.g. manifold support vector machine). However, the project is open to your ideas and whatever serves the purpose.
In advance it might be worth grasping some additional information:
- GutHub Repo of TES Tandem Tutorial "Population wide Medical Image and Shape Analysis" (Tutorial 2). Attention: The Mybinder server takes approx. 10min. to start, don't worry about it. Besides the code examples you can also find some slides.
- Morphomatics Library
- Update slides
- FAUST-data