Region-specific Diffeomorphic Metric Mapping
Authors: Zhengyang Shen, Francois-Xavier Vialard, Marc Niethammer
NeurIPS 2019 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | To evaluate RDMM performance we experiment 1) on synthetic 2D data and 2) on two 3D datasets: knee magnetic resonance images (MRIs) of the Osteoarthritis Initiative (OAI) and computed tomography images (CT) of the lung. Results show that our framework achieves comparable performance to state-of-the-art image registration approaches, while providing additional information via a learned spatio-temporal regularizer. Further, our deep learning approach allows for very fast RDMM and LDDMM estimations. |
| Researcher Affiliation | Academia | Zhengyang Shen UNC Chapel Hill zyshen@cs.unc.edu François-Xavier Vialard LIGM, UPEM francois-xavier.vialard@u-pem.fr Marc Niethammer UNC Chapel Hill mn@cs.unc.edu |
| Pseudocode | No | The paper presents mathematical formulations and theorems (e.g., Theorem 1, Theorem 2) but does not include any structured pseudocode or algorithm blocks. |
| Open Source Code | Yes | Code is available at https://github.com/uncbiag/registration. |
| Open Datasets | Yes | The OAI dataset consists of 176 manually labeled MRI from 88 patients (2 longitudinal scans per patient) and 22,950 unlabeled MR images from 2,444 patients. Labels are available for femoral and tibial cartilage. We divide the patients into training (2,800 pairs), validation (50 pairs) and testing groups (300 pairs), with the same evaluation settings as for the cross-subject experiments in [33]. |
| Dataset Splits | Yes | We divide the patients into training (2,800 pairs), validation (50 pairs) and testing groups (300 pairs), with the same evaluation settings as for the cross-subject experiments in [33]. |
| Hardware Specification | No | The paper does not provide specific details about the hardware used for experiments, such as GPU models, CPU types, or memory specifications. It does not mention any cloud or cluster resource specifications. |
| Software Dependencies | No | The paper mentions software components and algorithms like 'dopri5 solver', 'Adam', '3D UNets', and 'L-BGFS' but does not specify version numbers for any of these components, which is required for reproducibility. |
| Experiment Setup | Yes | We use 3D UNets [9] for momentum and pre-weight prediction. Both the affine and the non-parametric networks can be iterated to refine the prediction results. We use the dopri5 solver using the adjoint sensitivity method [7] to integrate the evolution equations in time. For solutions based on numerical optimization, we use a multi-scale strategy with L-BGFS [19] as the optimizer. We use Adam [17] for optimization. We use multi-kernel Localized Normalized Cross Correlation (mk LNCC) [33] for image similarity. Inside the lung a regularizer with small standard deviation (σi = {0.04, 0.06, 0.08} , h2 0 = {0.1, 0.4, 0.5}) and outside the lung with large standard deviation is used (σi = {0.2}, h2 0 = {1.0}). |