Benchmarking the selected configurations in:
Dalca, A.V., Balakrishnan, G., Guttag, J. and Sabuncu, M.R., 2019. Unsupervised learning of probabilistic diffeomorphic registration for images and surfaces. Medical image analysis, 57, pp.226-236.
This is related to #3.
Summary:
-
Tasks:
Unsupervised algorithms
Optional: surface-based registration (segmentation maps)
-
Transformation:
Stationary-ODE-based diffeomorphism by predicting velocity and integration using scaling-and-square integration (DVF in DeepReg).
-
Network and loss:
3D UNet starting with 32 filters; (although similar, maybe worth a re-implementation), outputting at 1/2 voxel size (equivalent to extract_levels: [3] in DeepReg)
difference using reparameterisation to predict the Gaussian parameters for the probabilistic loss; and the loss is minimising the lower bound, resulting in an image data term and a regularising term - interesting to see the difference to intra-SSD and inter-NCC.
-
Data:
atlas-based registration, i.e. register each image to an atlas computed independently
-
Metrics:
Dice on warped segmentation maps
Jacobian
Benchmarking the selected configurations in:
Dalca, A.V., Balakrishnan, G., Guttag, J. and Sabuncu, M.R., 2019. Unsupervised learning of probabilistic diffeomorphic registration for images and surfaces. Medical image analysis, 57, pp.226-236.
This is related to #3.
Summary:
Tasks:
Unsupervised algorithms
Optional: surface-based registration (segmentation maps)
Transformation:
Stationary-ODE-based diffeomorphism by predicting velocity and integration using scaling-and-square integration (DVF in DeepReg).
Network and loss:
3D UNet starting with 32 filters; (although similar, maybe worth a re-implementation), outputting at 1/2 voxel size (equivalent to extract_levels: [3] in DeepReg)
difference using reparameterisation to predict the Gaussian parameters for the probabilistic loss; and the loss is minimising the lower bound, resulting in an image data term and a regularising term - interesting to see the difference to intra-SSD and inter-NCC.
Data:
atlas-based registration, i.e. register each image to an atlas computed independently
Metrics:
Dice on warped segmentation maps
Jacobian