Title
MOSAIC – Motif-based Surface Alignment & Inference of Cortex
Short description and the goals for the OHBM BrainHack
Cortical folding emerges rapidly across the perinatal period, and its pattern varies strikingly between individuals. This variability is a problem for standard surface registration, which assumes a one-to-one correspondence between every subject and a single template and so tends to smooth away precisely the folding differences we want to measure. Motif-based approaches address this by matching each subject to the most similar pattern in a library of folding templates rather than forcing everyone onto one average, preserving individual morphology while still enabling group analysis. MOSAIC brings this idea to fetal and neonatal data, combining motif selection, label propagation, and feature extraction into a single pipeline whose outputs support normative modelling of how folding develops – and where individual brains deviate.
Given a new subject's hemisphere surface, MOSAIC aligns it to HCP space, registers it against a library of folding-motif templates, and selects the closest motif by Deep-Discrete spherical Registration (DDR) and multifacet similarity scores. Using Multimodel surface Matching (MSM), it propagates each template's segmentation and sulcal-fundus labels onto the subject, then extracts surface features (sulcal length, width, cortical thickness, folding extent) in native space. Outputs feed into normative modelling of folding morphometry across perinatal age. The hackathon focuses on label propagation, feature extraction, batch processing of the dHCP cohort, and wiring the feature table into the GP normative-modelling code – packaged as a runnable tool with example data.
Link to the Project
https://github.com/metrics-lab/MOSAIC
Image/Logo for the OHBM brainhack website
No response
Project lead
Yourong Guo, Yrong-Guo, yourong_g
Nashira Baena, palomanashira, palomanashira
Jiaxin Xiao, NNN-Bec, jiaxin0968
Kaili Liang, kaili23, kaili_liang
Shang Shi, sabrinaaaaas, sab0061
Emma Robinson, ecr05, emma_00188
Main Hub
Bordeaux
Link to the Project pitch
https://github.com/metrics-lab/MOSAIC/blob/main/README.md
Other hubs covered by the leaders
Skills
- Python and the surface-neuroimaging stack: nibabel, GIFTI handling, FreeSurfer and/or Connectome Workbench
- Comfort with surface registration concepts (spherical/diffeomorphic, MSM-style alignment)
- Mesh/geometry processing: numpy, scipy, and something like pyvista/vtk or nilearn for geodesic and curvature work (especially for feature-extraction)
- Label propagation on surfaces
- Normative modelling experience (basic Bayesian/GAMLSS regression)
- Git and light packaging hygiene
- Bonus: experience with neonatal/dHCP data and developmental neuroimaging
Recommended tutorials for new contributors
No response
Good first issues
No response
Twitter summary
No response
Short name for the Discord chat channel (~15 chars)
MOSAIC
Please read and follow the OHBM Code of Conduct
Title
MOSAIC – Motif-based Surface Alignment & Inference of Cortex
Short description and the goals for the OHBM BrainHack
Cortical folding emerges rapidly across the perinatal period, and its pattern varies strikingly between individuals. This variability is a problem for standard surface registration, which assumes a one-to-one correspondence between every subject and a single template and so tends to smooth away precisely the folding differences we want to measure. Motif-based approaches address this by matching each subject to the most similar pattern in a library of folding templates rather than forcing everyone onto one average, preserving individual morphology while still enabling group analysis. MOSAIC brings this idea to fetal and neonatal data, combining motif selection, label propagation, and feature extraction into a single pipeline whose outputs support normative modelling of how folding develops – and where individual brains deviate.
Given a new subject's hemisphere surface, MOSAIC aligns it to HCP space, registers it against a library of folding-motif templates, and selects the closest motif by Deep-Discrete spherical Registration (DDR) and multifacet similarity scores. Using Multimodel surface Matching (MSM), it propagates each template's segmentation and sulcal-fundus labels onto the subject, then extracts surface features (sulcal length, width, cortical thickness, folding extent) in native space. Outputs feed into normative modelling of folding morphometry across perinatal age. The hackathon focuses on label propagation, feature extraction, batch processing of the dHCP cohort, and wiring the feature table into the GP normative-modelling code – packaged as a runnable tool with example data.
Link to the Project
https://github.com/metrics-lab/MOSAIC
Image/Logo for the OHBM brainhack website
No response
Project lead
Yourong Guo, Yrong-Guo, yourong_g
Nashira Baena, palomanashira, palomanashira
Jiaxin Xiao, NNN-Bec, jiaxin0968
Kaili Liang, kaili23, kaili_liang
Shang Shi, sabrinaaaaas, sab0061
Emma Robinson, ecr05, emma_00188
Main Hub
Bordeaux
Link to the Project pitch
https://github.com/metrics-lab/MOSAIC/blob/main/README.md
Other hubs covered by the leaders
Skills
Recommended tutorials for new contributors
No response
Good first issues
No response
Twitter summary
No response
Short name for the Discord chat channel (~15 chars)
MOSAIC
Please read and follow the OHBM Code of Conduct