Patch2Self: Denoising Diffusion MRI with Self-Supervised Learning
Authors: Shreyas Fadnavis, Joshua Batson, Eleftherios Garyfallidis
NeurIPS 2020 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We demonstrate the effectiveness of Patch2Self via quantitative and qualitative improvements in microstructure modeling, tracking (via fiber bundle coherency) and model estimation relative to other unsupervised methods on real and simulated data. |
| Researcher Affiliation | Collaboration | 1Indiana University Bloomington, 2CZ Biohub |
| Pseudocode | Yes | Algorithm: Patch2Self |
| Open Source Code | No | To enable broad adoption of this method by the MRI community, we will incorporate an efficient and unit-tested implementation of Patch2Self into the widely-used open-source library DIPY. |
| Open Datasets | Yes | We compare the performance of Patch2Self with Marchenko-Pastur on the Parkinson s Progression Markers Initiative (PPMI) [32], Stanford HARDI [41] and Sherbrooke 3-Shell [16] datasets as shown in Fig. 2. |
| Dataset Splits | Yes | In order to compare the goodness of each fit, we perform a k-fold cross-validation (CV) [22] at two exemplary voxel locations, corpus callosum (CC), a single-fiber structure, and centrum semiovale (CSO), a crossing-fiber structure. The data is divided into k = 3 different subsets for the selected voxels, and data from two folds are used to fit the model, which predicts the data on the held-out fold. |
| Hardware Specification | No | The paper does not provide any specific details about the hardware used for running the experiments. |
| Software Dependencies | No | Our code-base allows for the use of any regression model from [39]. While scikit-learn is mentioned implicitly through the citation, no specific version numbers for scikit-learn or other software dependencies are provided. |
| Experiment Setup | Yes | In the remainder of the text, we use and show results with patch radius zero and linear regressors. To perform the probabilistic tracking, the data was first fitted with the Constant Solid Angle (CSA) model [1]. The Generalized Fractional Anisotropy (GFA) metric extracted from this fitting was used as a stopping criterion within the probabilistic tracking algorithm. The fiber orientation distribution information required to perform the tracking was obtained from the Constrained Spherical Deconvolution (CSD) [46] model fitted to the same data. ...a seeding density of 6. |