QSM-CI method: AFTER-QSM (PyTorch) (v2)
Year: 2026
Type: software
Abstract: AFTER-QSM (Affine Transformation Edited and Refined) — a pretrained, affine-equivariant deep neural network for QSM dipole inversion that is robust to ARBITRARY head orientation and ANISOTROPIC resolution (down to 0.6 mm isotropic), where axial-trained networks fail. It (1) affine-transforms the input local field into the network's canonical axial / 0.6 mm frame, (2) runs a U-net for coarse dipole inversion, (3) inverse-affine-transforms back, then (4) runs a Residual-Dense refinement network to deblur and correct susceptibility underestimation from the affine resampling. Consumes the local (tissue) field in ppm and produces susceptibility (ppm), on the input affine. The voxel size and B0 direction (z-projections) are passed to the affine stage from params. GPU-only in practice (8 GB+ VRAM); see the ci_skip note below. QSM-CI reconstruction method after-qsm. Browse and run it at https://qsmxt.github.io/QSM-CI/submission.html?method=after-qsm.
Language: English
Subfield: Radiology, Nuclear Medicine and Imaging
Field: Medicine
Domain: Health Sciences
Open Access status: green
OpenAlex ID: W7204189026