magetbrain

MAGeT-Brain: Multiple Automatically Generated Templates Brain Segmentation.

Performs multi-atlas segmentation through combinatorial registration and majority vote label fusion. The algorithm propagates atlas labels to subjects via an intermediate template library, leveraging neuroanatomical variability to improve segmentation accuracy.

Pipeline steps:
1. Register every atlas to every template (REGISTER_ATLAS_TEMPLATE)
2. Register every template to every subject (REGISTER_TEMPLATE_SUBJECT)
3. Chain transforms to resample atlas labels into subject space via every
possible atlas-template-subject path (RESAMPLE_LABELS)
4. Majority vote across all candidate labels per subject (IMAGE_ANTSMATH)

Computational note: With A atlases, T templates, S subjects, and L labels per atlas, the subworkflow produces AT + TS registrations and ATS*L label resamplings.

References:
- Pipitone J, Park MT, Winterburn J, et al. Multi-atlas segmentation of the
whole hippocampus and subfields using multiple automatically generated templates. Neuroimage. 2014.
- Chakravarty MM, Steadman P, van Eede MC, et al. Performing label-fusion-
based segmentation using multiple automatically generated templates. Hum Brain Mapp, 34(10):2635-54, October 2013. doi:10.1002/hbm.22092

Implementation assistance: Claude (Anthropic). Models: Opus 4.6, Opus 4.7

Keywords : segmentation, multi-atlas, label fusion, majority voting, MAGeTBrain, registration

Components : registration/ants, registration/antsapplytransforms, image/antsmath

---

Inputs

	Type	Description	Mandatory	Pattern
ch_atlases	file	Channel containing atlas brain images. Each atlas may have one or more associated label files provided via ch_labels. Structure : tuple val(meta), path(image) meta [map] Metadata map. Must contain id (atlas identifier). e.g. [ id:'atlas1' ] image [file] Atlas brain image.	True	*.\{nii,nii.gz\}
ch_labels	file	Channel containing atlas label (segmentation) files. The meta.id must match the corresponding atlas, and meta.label_tag identifies the label type (e.g. “_label_amygdala”). Structure : tuple val(meta), path(label) meta [map] Metadata map. Must contain id (atlas identifier matching ch_atlases) and label_tag (label type identifier, e.g. “_label_amygdala”). e.g. [ id:'atlas1', label_tag:'_label_amygdala' ] label [file] Atlas label segmentation file.	True	*.\{nii,nii.gz\}
ch_templates	file	Channel containing template brain images. Templates serve as intermediate registration targets between atlases and subjects. Structure : tuple val(meta), path(image) meta [map] Metadata map. Must contain id (template identifier). e.g. [ id:'template1' ] image [file] Template brain image.	True	*.\{nii,nii.gz\}
ch_subjects	file	Channel containing subject brain images to be segmented. Structure : tuple val(meta), path(image) meta [map] Metadata map. Must contain id (subject identifier). e.g. [ id:'subject1' ] image [file] Subject brain image.	True	*.\{nii,nii.gz\}

---

Outputs

	Type	Description	Mandatory	Pattern
labels	file	Channel containing majority-voted segmentation labels for each subject and label type. Structure : tuple val(meta), path(label) meta [map] Metadata map containing id (composed of subject identifier and label tag, e.g. “subject1_label_amygdala”). label [file] Majority-voted label segmentation file.	True	*_majorityvote.nii.gz
versions	file	File containing software versions used. Structure : tuple path(versions) versions [file] Versions YAML file.	True	versions.yml

---

Parameters (see configuration)

	Type	Description	Default	Choices
single_thread	boolean	{‘type’: ‘boolean’, ‘description’: ‘Force single-threaded execution for registration and resampling processes.’, ‘default’: False}	False
repro_mode	boolean	{‘type’: ‘boolean’, ‘description’: ‘Run registration in reproducibility mode (registration/ants only).’, ‘default’: False}	False

---

Authors

• @CMonnin

---

Last updated : 2026-05-12