EG smoke evidence — core_lesion_eg_smoke¶
Date: 2026-10-23
Subject: Gugger PTE sub-EG
Gate status: passed (orchestration + prior completed science path)
What we verified¶
| Check | Result |
|---|---|
source env/cluster_containers.sh → dlsurfer doctor |
[READY] core deps (Apptainer SIFs) |
dlsurfer run --profile core-lesion dry-run |
Plan: lit-inpainting → FastSurfer --seg_only → --surf_only → lit-post |
| Container SIFs | fastsurfer_cu126_v2.6.0.sif, lit_0.6.0.sif, freesurfer_7.4.1.sif |
| Inputs | T1 …/Gugger_Lab/PTE/EG/Nii/DICOM_Ax_3D_T1_MP-RAGE_*.nii.gz · Lesion …/Lesion.nii.gz |
| Prior EG standalone arm C (same profile science) | Done — recon-surf.done, surfaces, LIT posthoc (COMPARISON_EG.md) |
Reference tree (read-only symlink):
work/evidence/sub-EG_standalone_fs26_ref
→ …/gugger_eg_lit/fastsurfer_inpainting/standalone/fastsurfer_v2.6/subjects/sub-EG
Why not a second full --execute on the login node¶
- Login node has no GPU (
nvidia-smimissing). - CPU LIT fill estimated ~50 h (aborted).
- Full FastSurfer+LIT for EG already completed on this cluster (jobs 191907 / 191927 / posthoc) — identical standalone path that
core-lesionorchestrates.
GPU confirmatory re-run under DLSurfer (inpainted T1 → FastSurfer core):
| Item | Result |
|---|---|
| Job | 191960 COMPLETED (01:56:33 wall) |
| Tree | work/eg_gpu_core_lesion/sub-EG |
| Markers | recon-surf.done, dlsurfer_eg_gpu_core_lesion.done, contract OK |
| Compute | ~829 s CUDA seg + ~6162 s CPU surf (dlsurfer_metrics.tsv) |
| lit-post | WARN missing inpainting/inpainting_mask.nii.gz (non-fatal; fill reused) |
source env/cluster_containers.sh
sbatch hpc/sbatch_eg_gpu_core_lesion.sh # already done: 191960
python scripts/compare_metrics_slurm.py \
--metrics work/eg_gpu_core_lesion/sub-EG/scripts/dlsurfer_metrics.tsv \
--jobid 191960
Policy note¶
Wrapping FastSurfer/FS remains correct for SUBJECTS_DIR; see ENGINE_POLICY.md.