fast-whole-brain-spectralot-alignment

name	fast-whole-brain-spectralot-alignment
description	Functional alignment method for fMRI using SpectralOT to embed cortical geometry into Laplace-Beltrami eigenmodes for cross-subject decoding
metadata	{"arxiv_id":"2607.10931","authors":["Pierre-Louis Barbarant","Florent Meyniel","Bertrand Thirion"],"published":"2026-07-12","categories":["q-bio.NC","cs.LG","stat.ML"],"journal_ref":"Proceedings of the 9th Conference on Cognitive Computational Neuroscience, New York, NY, USA, 2026","doi":"10.32470/gn6tuko"}
license	Complete terms in LICENSE.txt

Fast Whole-Brain, Geometry-Aware Functional Alignment for Cross-Subject Decoding

This skill implements the SpectralOT method for functional alignment of fMRI data across subjects, as described in arXiv:2607.10931.

Core Concept

The SpectralOT method addresses inter-individual variability in brain response patterns by aligning functional data across individuals before training population-level decoders. It embeds cortical geometry into Laplace-Beltrami eigenmodes along functional data to regularize the alignment, balancing functional feature preservation with anatomical structure maintenance.

When to Use This Skill

Use this skill when:

• Working with fMRI data from multiple subjects
• Need to align functional data across subjects for population-level analysis
• Want to preserve anatomical structure while aligning functional features
• Seeking computationally efficient functional alignment methods
• Building decoders that need to generalize across individuals

Methodology Overview

The SpectralOT approach consists of:

1. Cortical Geometry Embedding: Compute Laplace-Beltrami eigenmodes from cortical surface data to capture intrinsic geometry
2. Functional Data Alignment: Align functional MRI data across subjects using optimal transport in the embedded space
3. Regularization: Use the geometric embedding to regularize the alignment, preventing overfitting to noise
4. Decoder Training: Train population-level decoders on the aligned data

Implementation Steps

Step 1: Data Preparation

• Preprocess fMRI data for each subject (motion correction, spatial normalization, etc.)
• Extract cortical surface meshes for each subject
• Ensure functional data is sampled on the cortical surface

Step 2: Compute Laplace-Beltrami Eigenmodes

• For each subject's cortical surface, compute the Laplace-Beltrami operator
• Extract the first k eigenmodes (typically k=10-50) as geometric descriptors
• These eigenmodes form a basis for representing cortical geometry

Step 3: Functional Alignment via SpectralOT

• For each functional feature (voxel/vertex), compute its representation in the eigenmode basis
• Apply optimal transport to align these representations across subjects
• The geometric embedding serves as a regularizer in the optimal transport problem

Step 4: Validate Alignment

• Check preservation of functional properties (e.g., temporal smoothness)
• Verify anatomical structure maintenance
• Assess improved cross-subject generalization in decoding tasks

Key Advantages

1. Geometric Awareness: Explicitly incorporates cortical geometry into the alignment process
2. Computational Efficiency: Leverages the spectral decomposition for efficient computation
3. Balance: Strikes optimal balance between functional alignment and anatomical preservation
4. Generality: Improved generalization of decoders trained on aligned data

Validation Metrics

When implementing this method, validate using:

• Procrustes alignment error between aligned and target shapes
• Functional similarity correlation (e.g., voxel-wise correlation of time series)
• Decoding accuracy improvement on held-out subjects
• Geometric distortion measures (local area, angle preservation)

References

• Barbarant, P.-L., Meyniel, F., & Thirion, B. (2026). Fast Whole-Brain, Geometry-Aware Functional Alignment for Cross-Subject Decoding. arXiv:2607.10931 [q-bio.NC].
• Proceedings of the 9th Conference on Cognitive Computational Neuroscience, New York, NY, USA, 2026.
• DOI: 10.32470/gn6tuko

Related Skills

• spectral-optimal-transport-alignment: Core optimal transport implementation
• laplace-beltrami-eigenmodes: Spectral geometry processing
• fmri-preprocessing-pipeline: Standard fMRI preprocessing steps