All credible alternatives (chosen + runners-up)¶
Policy. Each stage has one chosen default in engines.yaml (chosen: true). All other credible options stay documented here and under alternatives: / deferred: for future swaps — without pulling them into the default path or into DKT connectome.
Readiness: P Production · R Ready · S Research · E Early · C Classical kept
| Stage | Chosen | Other best alternatives (keep for later) | Why not default |
|---|---|---|---|
| Skull-strip | SynthStrip (P) | HD-BET (R); ROBEX (classic ML); nnU-Net/MONAI extractors (S) | SynthStrip already in FS; one-license path |
| Bias field | DeepN4 (R) | Simkó implicit CNN (S); Log-Gabor CNN bias (S); N4ITK classical (C) | DeepN4 portable + public weights; optional util only |
| Whole-brain seg | FastSurferVINN (P) | SynthSeg (P); QuickNAT (R/S); DeepSCAN (R/S); AssemblyNet (S); SLANT (S) | VINN = DKT-native + hi-res; SynthSeg kept as odd-contrast profile |
| CC | FastSurfer-CC (P) | Classical FS CC post (C) | Upstream FastSurfer default |
| Cerebellum | CerebNet (P) | Classical / atlas cereb (C) | Upstream module |
| Hypothalamus | HypVINN (P) | ScLimbic (P, FS); other subunit nets (S) | XOR: HypVINN with FastSurfer engine |
| Lesion fill | neuroLIT standalone (P) | Native --lesion_mask (P); VBG (R); LeAPP VBT (R); enantiomorphic (C/R) |
Standalone keeps GPU/CPU split; VBG mixed on TBI; VBT TBI crash risk |
| Surfaces (default) | recon-surf (C) | — | Still best SUBJECTS_DIR contract |
| White surface DL | TopoFit (P) | V2C-Flow (R); Vox2Cortex (R); CortexODE (S); CorticalFlow (S); SurfNN (S); DeepCSR (S); FastCSR (S) | TopoFit ships in FS; others need TBI±LIT gate |
| Pial / joint CSR | V2C-Flow (R) (Phase 3) | PialNN (S); CortexODE (S); SurfNN (S); Vox2Cortex (R); CorticalFlow (S); recon-all-clinical SynthDist (P hybrid) | V2C-Flow = best Ready full WM+pial + template IDs |
| sphere.reg | recon-surf spectral/FS (C); SUGAR (R) Phase 3 | S3Reg (R); V2C-Flow baked-in IDs (R) | Classical default; SUGAR if reg is bottleneck |
| Volume registration | EasyReg + SynthMorph (P) | VoxelMorph (S); TransMorph (S); ANTs (C) | Both ship in FS |
| Thickness w/o mesh | DL-CTE family (R) | DL+DiReCT (R); CortexMorph (R); direct CNN maps (S) | Biomarker only — never default morphometry |
| Alt full engine | FreeSurfer 8 (P) | FastSurfer core (P) |
XOR per subject — never mix mid-tree |
| Longitudinal | FS -base/-long (P) via fs-long |
SynthSeg-per-TP (P partial); V2C-Long (E) | Classical long is still best |
| Subfields HA | segmentHA_T1 (P) via subfields-ha |
HippUnfold (R); Hippodeep (S) | Wired gap-fill |
| Brainstem / thalamic | segmentBS + ThalamicNuclei (P) via nuclei-bs-thal |
(none stronger for FS-tree) | Wired gap-fill |
| Clinical / odd contrast | recon-all-clinical (P) via fs-clinical |
SynthSeg volume; Infant FS (neonates) | Infant FS stays external |
| Cortical DK | --fsaparc / core-dk (P) |
FS8 aparc | Wired |
| Destrieux | -a2009s / atlas-destrieux (P) |
mris_ca_label |
Wired |
| Schaefer / MMP | atlas-schaefer surf2surf (R) |
HCP-MMP / Yeo same pattern | Wired |
| HPC launcher | --hpc srun\|brun (P) |
hpc/slurm_core_lesion.sh |
Wired |
| Hires VINN | --vox-size (P) |
— | Wired |
| Module off switches | --no-cc/cereb/hypothal (P) |
— | Wired |
| Claustrum | Mauri mri_claustrum_seg (P) via structure-claustrum |
Albishri CNN; Casamitjana Bayesian | Lab Documents/clau_seg |
| Piriform (125/126) | SynthSeg-PC (R/E) via structure-piriform-synthseg |
JLF (P, IDEAS now); MAPER (P); nnU-Net PC (S) | Gate before production |
| Clinical T1 salvage | SynthSR (P) via preprocess-synthsr |
skip; fs-clinical |
Preprocess only |
| HA unfold morphometry | HippUnfold (R) via structure-hippunfold |
segmentHA (default); Hippodeep | Lab Documents/HippUnfold |
Gaps with no credible alternative: GAPS_WITHOUT_ALTERNATIVES.md.
Specialty wiring: STRUCTURE_MODULES.md.
Citations for runners-up (quick)¶
- HD-BET — Isensee et al. (brain extraction)
- QuickNAT — Roy et al.
- DeepSCAN — McKinley et al.
- VBG — Radwan et al. NeuroImage 2021
- LeAPP / VBT — Bey et al. HBM 2024
- Enantiomorphic — Nachev et al.; Fekonja et al. 2022 (with FastSurfer)
- CortexODE — Ma et al. TMI 2022
- PialNN — Ma et al. 2021
- CorticalFlow — Lebrat et al. 2022
- SurfNN — Zheng et al. ISBI 2023
- DeepCSR — Santa Cruz et al. WACV 2021
- FastCSR — Ren et al. Brain Informatics 2022
- S3Reg — Zhao et al. TMI 2021
- Deutscher et al. 2025 — VBG caution on ms-TBI
Machine-readable mirrors: engines.yaml → alternatives: on each chosen engine + top-level deferred:.