TopCoW — Worker pipeline v2

Cerebral vessel segmentation (14 classes) — all compute (WASM + WebGPU) runs in a dedicated Web Worker so the page stays fully responsive during the ~75s inference. Strict nnU-Net V2 pipeline: transpose → crop-to-nonzero → per-image ZScore → cubic B-spline resample → sliding window. · ← v1 (main-thread)

✓ nnU-Net V2 strict conformance: canonical-frame argmax 99.9989% vs nnUNetv2_predict (fold 0, use_mirroring=False, use_gaussian=True). Chrome WebGPU ≈ Python MPS (~75 s on M1, 180 patches).

② Upload NIfTI (.nii / .nii.gz)

Log

Axial slice viewer

Slice z: 0

Canonical input (Z-score normalized)

Segmentation

Overlay