Week of 2026-07-06

syn7T

Aim: work out why coregistration was failing and settle the 7T ground-truth sequence.

Reconstructed the pipeline order from the submitted batch scripts and found a label-swapping bug in compute_metrics.py — the two BME-X arms (BME-X-then-coreg vs. coreg-then-BME-X) had their dictionary keys reversed. Separately, syn7T/raw_nifti is stale; the real data lives at syn7T/data/raw_nifti. BIDS filenames carry acq-3T/acq-7T entities that many batch scripts don’t account for, so some subjects may have been silently skipped.

A cohort audit script surfaced real protocol heterogeneity: seven subjects (sub-031/032/033/034/036/037/038) have plain 7T MPRAGE rather than MP2RAGE, three already have INV2, sub-044 is missing 3T entirely, and 3T protocols span roughly 25 distinct SeriesDescriptions. sub-026 is confirmed post-op on visual inspection; sub-008 shares the same suspicious coronal high-resolution geometry (512x176x512) and needs checking.

Rather than pick between MP2RAGE UNI-DEN and MPRAGE as ground truth before downloading, I decided to pull all four 7T sequences (UNI-DEN/UNIT1, INV1, INV2, clinical MPRAGE) for every subject and defer the choice. Dr. Jones confirmed MP2RAGE is used for research and MPRAGE for clinical, noted the B1-correction advantage of MP2RAGE is less critical with PTX hardware, and softly suggested clinical MPRAGE for consistency without committing. I also changed the brain-extraction strategy: run SynthStrip on INV2 (cleaner background) and apply that mask to UNIT1, rather than running SynthStrip on the noisy UNI-DEN background directly — this follows published MP2RAGE practice. Revised pipeline order: download all four sequences, dcm2bids with the updated config, SynthStrip on INV2, apply the INV2 mask to UNIT1, N4 bias correction, registration comparison arms. Scripts produced: build_scan_audit.py, batch_synthstrip_7T_INV2.sh, batch_apply_inv2_mask_to_unit1.sh, dcm2bids_config_additions.json, syn7T_preprocessing_tracker.xlsx.

Later in the week the download finished — all four 7T sequences for the 52-subject cohort are off PACS. Two MP2RAGE wrinkles came up. One patient had INV1, INV2, and UNI but no UNI-DEN; applying an INV2-derived mask to UNI is not equivalent to UNI-DEN, which uses a regularized recombination formula (O’Brien 2014 / Marques 2010) that suppresses noise amplification in low-SNR regions throughout the brain, not just the background. The fix is to regenerate a true UNI-DEN with RobustCombination.m or LAYNII’s LN_MP2RAGE_DNOISE (beta=0.2). Another patient had four identically labelled cs tfl_w ip925b_0.63mm_sag series, almost certainly the four outputs of a compressed-sensing WIP MP2RAGE protocol; InversionTime (0018,0082) and ImageType (0008,0008) separate INV1/INV2 from the derived pair, and SeriesNumber ordering or visual inspection of low-B1 regions separates UNI from UNI-DEN.

Next: dry-run dcm2bids on sub-001 before any full-cohort run, verify sub-008, and confirm with Steve Jones whether the WIP protocol string orders its outputs consistently across subjects.

MRF-SEEG

Aim: get resection masks coregistered so each electrode contact can be labelled resected vs. non-resected.

Confirmed the target space first. The FreeSurfer-conformed space (conv_T1_fs, 256^3 LIA) and the native res_T1_bet space (210x288x210, RAS) are distinct grids serving different purposes. Electrode coordinates and all MRF sampling scripts operate on the res_T1_bet grid, so that’s the registration target, not FS space.

Ran the four-step pipeline on ith2: brain extraction with mri_synthstrip; valid-tissue mask via fslmaths (brain mask minus dilated resection cavity, so the post-op cavity is excluded from the ANTs cost function); masked SyN registration of the post-op brain to res_T1_bet with antsRegistrationSyN.sh; and antsApplyTransforms with NearestNeighbor interpolation to bring the binary resection mask into res_T1_bet space. Visual QC confirmed correct anatomical placement. Two environment gotchas worth recording: res_T1_bet_mask.nii lives in ANTs_1mm/, not the patient root, and fslmaths needed export PATH=$FSLDIR/bin:$PATH despite $FSLDIR being set. Wrote a batch script for the remaining nine patients (ith3, 4, 5, 6, 8, 9, 11, 12, 13) that handles both resection-mask naming conventions (_postop_lacuna.nii.gz for ith11/ith13 vs. _postop_lacuna_lacuna.nii.gz for the rest), with per-step idempotency checks and timestamped logging.

Also tracked down where the VEP-normalized MRF brains live for electrode-level sampling: \\thorb.ccf.org\eegrvw-h\Imaging\Multimodal\MRF_SEEG_Max\pilot\ith{N}\ANTs_1mm\. A normalization comparison batch sits at .../Research/MRF-SEEG/data-raw/2025-11-04-normalized/ith{N}/ with variants named raw, fcm, quantile, zscore_robust, zscore_whole, zscore_wm, and zscore_tissue. 2025-11-24-generate-vep-normalized-brains.R is the VEP z-score pipeline: a weighted GM/WM pooled z-score using per-region normative statistics from data/2025-06-27-vep-lookup-table.csv and VEP atlas labels from aparc+aseg.vep.nii.gz.

Next: VEP region labelling per contact would need a coordinate transform out of res_T1_bet (flagged as future work), and it’s still unresolved whether zscore_tissue is the actual VEP-normalized output or a simpler global tissue normalization — a verification script would settle it.

HEMI-SLIM

Verbally consented the first family by phone, reached the second, and sent both the journal’s official consent form via DocuSign.

Reviewed the ESR-FLAIR FCD detection paper (submission WNL-2026-303475) and flagged: a patient-count discrepancy in the Table 2 seizure-outcome row (66+18=84, not 93), a citation error labelling MAP as “MAP18” when reference 18 is the SynthSR paper rather than Huppertz et al. (reference 43), duplicate references 13 and 14, rounding inconsistencies between text and tables, an apparently misplaced reference 40, an abstract word-count overage, a blank Author Contributions line, and an incomplete reference 46. Sent to Spencer and Irene. Separately, a Neurology editorial message came in requiring consent-form corrections for two patients, with specific instructions about incomplete and conflicting signature sections.

FB-GEN

Reviewed Bijoya’s draft (FBGEN_Project_ManuscriptDraft_v5.docx) on caregiver social media use and healthcare utilization in pediatric epilepsy. Issues found: a broken sentence in the Discussion, an incomplete exclusion-criteria sentence in Methods, a discrepancy between the engagement-group subtotals (166) and the analytic cohort (228), a gender reporting gap, a mismatch between Pearson’s correlation in Methods and Spearman’s rho in Results, unfilled placeholders (citations, sample sizes, R version, funding), an unclosed quotation mark, and an unspecified statistical test for group comparisons.

Sourced three thematic citation additions and drafted them as paragraph-level insertions for the Introduction and Discussion: caregiver and patient use of large language models (Yang et al. 2024, Epilepsy & Behavior; Kim et al. 2023, Seizure), a Facebook-based rare-disease study expanding phenotypic understanding (Talaba et al. 2026, Pediatric Neurology), and short-form video misinformation (St-Pierre et al. 2026, Epileptic Disorders; Dedic et al. on TikTok epilepsy misinformation). These went in as tracked comments and were output as FBGEN_Manuscript_v5_with_suggested_citations.docx.

Next: print with markup, annotate, and email Bijoya.

TELE-PET

Identified the CTSC AI Themed Pilot mechanism (deadline August 17) as a target for TELE-PET: a machine-learning classifier over MRI, PET, EEG, and clinical variables to detect temporal encephaloceles and determine whether an encephalocele is the true EZ or incidental. Target cohort roughly N=50, actual N unverified. Structural requirements: two CTSC institutions or one plus a community partner, mandatory SPARC consultation before InfoReady submission, IRB approval by December 1 2026, no indirect costs, no equipment, no travel, up to $20k/year technical support and up to $5k PI salary contingent on no active NIH support as PI or co-PI. Raymond F. Muzic Jr. (Director, Quantitative Imaging Laboratory, UH/CWRU) is the strongest co-I candidate, with Zhenghong Lee secondary.

Next: the existing MULTI-PET/TELE-PET umbrella IRB may be broad enough to cover this cohort, which would accelerate the timeline considerably.