Week of 2026-06-29
MRF-SEEG
Aim: get post-op imaging and resection cavities into a usable state for contact-level resection labelling.
Completed post-op DICOM-to-NIfTI conversion for ith2-6, 8, 9, and 12 using dcm2niix (at /usr/local/fsl/bin/dcm2niix). ith11 and ith13 needed a PACS re-download, then converted without issue. Installed ResectVol DL in a conda environment (resectvol_dl) and ran it to generate resection-cavity masks; several will need manual refinement in ITK-SNAP.
Wrote pom_to_nifti.py to convert Curry .pom electrode-contact files into NIfTI label volumes for ITK-SNAP. The key finding: all_1mm.pom coordinates are voxel indices, not world-space millimetres — confirmed by a range check (X: 107-180, Y: 81-177, Z: 79-170), all inside the 210x288x210 grid.
A systematic audit cleared up the lingering coordinate-space confusion. The per-patient anchor space is res_T1_bet.nii (210x288x210, 1mm isotropic). The ANTs_1mm/mrf_2_t1_* files share an identical affine, and the z-score images (T1_z_ptspace10.nii.gz) share the same grid, which validates the existing electrode-stats extraction pipeline.
Three new directions came out of lab meeting: expand the cohort to all patients regardless of seizure freedom, stratify epileptogenicity analysis by pathology (focus on nonlesional and FCD), and integrate Pris’s streamline-based cortical depth-sampling method.
Also settled the analysis design for comparing MRF signal inside vs. outside the resection zone. The methodological question is a gross inside/outside comparison vs. a concentric ring/shell approach built on a signed distance transform from the resection surface. Separating GM and WM matters — mixing them buries the signal. Heterogeneity between open resections and LITT ablations argues for normalizing distance by cavity size, and a buffer zone at the resection boundary should be excluded to limit partial volume and registration error. The existing pipeline already carries FSL FAST PVE maps (PVE0/1/2 = CSF/GM/WM) warped into FS space, which handles tissue-class separation without new segmentation work. One clarification worth recording: applying the transform to the mask with nearest-neighbor interpolation is the right mental model; the cavity-masked registration is a separate idea, where the cavity is excluded from the ANTs cost function during the T1-to-T1 registration itself, not from the subsequent mask warping. Naming is now cleaned up: post-op T1s are {patient}_postop.nii.gz, masks are {patient}_postop_lacuna.nii.gz, both under post-op-nifti/{patient}/. Ran a single-patient mri_synthstrip test on ith2 before batching.
Next: batch brain extraction across all ten patients, then register post-op T1s and resection masks into the anchor space via ANTs rigid registration.
HEMI-SLIM
Aim: get the manuscript out the door to Neurology.
Submitted. Prep included reformatting the abstract to Neurology’s structured requirements, drafting and revising the cover letter, running a STROBE compliance check, and nominating reviewers (Gaillard, Guerrini, Ramantani). Neurology then requested patient consent for the MRI and PET figures, so I drafted a family-facing consent request explaining the publication context, confirming the images are de-identified, and making clear that declining has no effect on care.
Next: await the editorial response, and confirm whether Cleveland Clinic has a standard publication release form that should accompany the consent request (check template requirements with the IRB coordinator).
MULTI-PET / TELE-PET
Drafted a full IRB application for MULTI-PET, an umbrella quantitative FDG-PET protocol for pediatric epilepsy, with HEMI-SLIM v2 as the first sub-study. Fixed the two project names here: MULTI-PET (multilobar resections and PET) and TELE-PET (temporal encephaloceles and FDG-PET).