Drop an image anywhere, or choose an example.
Drop an image anywhere, or choose an example.
Dwidenoise removes noise from diffusion MRI in this browser with Patch2Self (Fadnavis et al. 2020), a self-supervised method that needs no noise model. Each volume is predicted by a linear regression on the other volumes of its group, b0 or diffusion-weighted, fitted over all voxels; noise that is independent between volumes cannot be predicted and so is removed. It is a port of DIPY 1.12.1 Patch2Self version 1 (patch radius 0, ordinary least squares); on a 137-volume test slab its float32 output equals DIPY's bit for bit, with DIPY's in-place centring defect corrected. There is no mask input: as in DIPY, every voxel is fitted. Denoise first, before Gibbs-ringing removal and before registration interpolates the data. The bval is required; a bvec and sidecar chosen with the image are passed through under the output's name. A dropped BIDS folder gives each DWI run with its bval, bvec and inherited JSON sidecar; choose the run when there is more than one.
Images are processed locally in this browser and are never uploaded.