deface

Remove facial features from a brain MRI for anonymization — entirely in your browser. No data is sent to any server.

Defacing runs niimath, which registers a bundled MNI template (affine, BSD) and zeros the face voxels:

• allineate (fast) — fast affine registration engine (Hellinger MI, with a robust Hellinger fallback) — the default
• allineate (fast, robustfov) — same fast engine after a -robustfov crop of the neck/inferior slices for a tighter face mask
• allineate (Hellinger) — the exhaustive Hellinger engine (-cost hel): slower, reference-quality fit (also offered with a -robustfov crop)
• reface / rescalp — instead of zeroing the face, composite a synthetic surface onto it (-reface, AFNI afni_refacer2-style): the subject is registered to a template and a template-space shell is back-projected onto the original grid. reface replaces the face; rescalp replaces the whole scalp. Both are also offered with a -robustfov crop (which only tightens the registration — the output keeps the input's full extent). T1-weighted images only: the shell supplies T1-like intensities, so compositing it onto a T2/FLAIR/PD/CT scan yields a surface that does not match the host image — use an allineate or mindgrab method for those
• mindgrab — deep-learning brain extraction (skull-strip): the brainchop MeshNet model masks out everything but the brain, running in your browser via @brainchop/mindgrab — on WebGPU where available, otherwise WebGL2

Visualization uses NiiVue; DICOM import uses dcm2niix.

Defacing uses the BSD-2 build of niimath compiled to WebAssembly and bundled with this app, so it is licensed BSD-2-Clause · thread-deterministic registration, fails closed on out-of-memory. Brain extraction uses @brainchop/mindgrab (MIT). Requires a browser with WebGPU (recent desktop Chrome, Edge, or Safari).

MindGrab can't run here

The mindgrab brain-extraction method runs a neural network in the browser. It uses WebGPU where that is available, falls back to WebGL2, and none of them can start on this browser.

Use an allineate method instead — they run in WebAssembly and need no GPU.