TMS targets in about a minute from routine clinical MRI: a GPU head-model route checked against SimNIBS charm on 1 mm and 5 mm scans

Background: MRI-guided TMS targeting needs an individual head model, about an hour of computing from a near-isotropic T1, while routine clinical MRI often has 5 mm slices. Objective: To test whether a GPU route reproduces SimNIBS charm's TMS targets, EEG positions and fiducials within a minute on an 8 GB GPU, from 1 mm and 5 mm T1 input. Methods: A segmentation network and a learned MNI warp were trained on charm's output for 548 heads. Under rules sealed before training, the route was compared with charm on the original 1 mm T1 in 100 held-out IXI heads (1 mm and three simulated 5 mm inputs) and 100 OASIS-3 heads from an unseen source, at ten MNI targets, thirteen rTMS EEG positions and four fiducials, and against three simpler routes. Results: Mean target error was 1.32 mm with 1 mm and 1.35 mm with 5 mm input in IXI, and 1.31 and 1.33 mm in OASIS-3; EEG positions and fiducials were within 1.76 mm; charm's own scan–rescan difference was 0.58 mm. One head took 35 s at a GPU peak of 7.1–7.5 GB. The pre-set 5 mm worst-point limit was not met. The route beat charm run directly on 5 mm scans on every endpoint; SynthSR followed by charm was 0.20 mm closer at the targets but 0.80–2.69 mm worse on the scalp, at 68 min per head. Conclusions: From a routine 5 mm T1, the route gives TMS targets and scalp positions within about 1.76 mm of charm, in under a minute on an 8 GB GPU. Single points can be off by more than 5 mm and should be checked. Preprint. Code: https://github.com/s3644/tms-headmodel (GPL-3.0). Trained model: doi:10.5281/zenodo.24215664 (CC BY-NC 4.0).

Authors

Institutions

Publication Details

Journal
Zenodo (CERN European Organization for Nuclear Research)
Published
2026-11-14
DOI
https://doi.org/10.5281/zenodo.24215659
Primary Topic
Transcranial Magnetic Stimulation Studies
Type
preprint
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
preprint

TMS targets in about a minute from routine clinical MRI: a GPU head-model route checked against SimNIBS charm on 1 mm and 5 mm scans

Jukrapope Jitpimolmard
Zenodo (CERN European Organization for Nuclear Research)
Transcranial Magnetic Stimulation Studies
preprint

TMS targets in about a minute from routine clinical MRI: a GPU head-model route checked against SimNIBS charm on 1 mm and 5 mm scans

Jukrapope Jitpimolmard
preprint en

Abstract

Background: MRI-guided TMS targeting needs an individual head model, about an hour of computing from a near-isotropic T1, while routine clinical MRI often has 5 mm slices. Objective: To test whether a GPU route reproduces SimNIBS charm's TMS targets, EEG positions and fiducials within a minute on an 8 GB GPU, from 1 mm and 5 mm T1 input. Methods: A segmentation network and a learned MNI warp were trained on charm's output for 548 heads. Under rules sealed before training, the route was compared with charm on the original 1 mm T1 in 100 held-out IXI heads (1 mm and three simulated 5 mm inputs) and 100 OASIS-3 heads from an unseen source, at ten MNI targets, thirteen rTMS EEG positions and four fiducials, and against three simpler routes. Results: Mean target error was 1.32 mm with 1 mm and 1.35 mm with 5 mm input in IXI, and 1.31 and 1.33 mm in OASIS-3; EEG positions and fiducials were within 1.76 mm; charm's own scan–rescan difference was 0.58 mm. One head took 35 s at a GPU peak of 7.1–7.5 GB. The pre-set 5 mm worst-point limit was not met. The route beat charm run directly on 5 mm scans on every endpoint; SynthSR followed by charm was 0.20 mm closer at the targets but 0.80–2.69 mm worse on the scalp, at 68 min per head. Conclusions: From a routine 5 mm T1, the route gives TMS targets and scalp positions within about 1.76 mm of charm, in under a minute on an 8 GB GPU. Single points can be off by more than 5 mm and should be checked. Preprint. Code: https://github.com/s3644/tms-headmodel (GPL-3.0). Trained model: doi:10.5281/zenodo.24215664 (CC BY-NC 4.0).

Zenodo (CERN European Organization for Nuclear Research)
Khon Kaen University (TH)
Transcranial Magnetic Stimulation Studies
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon