ManifoldNet Paper (published in the International conf. on IPMI 2019) Implementation for Diffusion MRI (dMRI)
Please cite the following papers if you use this code:
Chakraborty, R., Bouza, J., Manton, J., & Vemuri, B. C. (2020). Manifoldnet: A deep neural network for manifold-valued data with applications. IEEE Transactions on Pattern Analysis and Machine Intelligence.
Chakraborty R., Bouza J., Manton J., Vemuri B.C. (2019) A Deep Neural Network for Manifold-Valued Data with Applications to Neuroimaging. In: Chung A., Gee J., Yushkevich P., Bao S. (eds), Proceedings of the International Conference on Information Processing in Medical Imaging. (IPMI) 2019.
We are not currently including our in-house dataset. For this reason you will need to include your own DTI (Diffusion Tensor Image) dataset and modify the dataloader defined in model.py accordingly.
Once the data is ready, simply run train.py. Included under data/example.npy is an example dataset with 4 synthetically generated SPD-valued images. By default, running train.py will overfit a model to these samples, you will see training loss approaching 0.