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

2025-09-01 · Source: stat.ML updates on arXiv.org · Field: Science & Research — Health & Medical Research, Artificial Intelligence & Machine Learning, Mathematics & Computational Sciences · Depth: Expert, extended

Summary

SpectralOT, a new functional alignment method for fMRI, addresses inter-individual variability in brain response patterns to improve population-level decoders. This method embeds cortical geometry into Laplace-Beltrami eigenmodes along functional data, regularizing the alignment process. SpectralOT computes geometric and functional dissimilarity matrices, interpolating them with a single parameter, alpha, before using a Sinkhorn Entropic Optimal Transport solver. Experiments demonstrate that SpectralOT consistently outperforms the anatomical baseline and the ProMises model in Inter-Subject Pearson Correlation (ISC) and cross-subject decoding. It also achieves statistically equivalent decoding performance to FUGW while being 30.5x faster, completing alignment and test data projection in 33.55s compared to FUGW's 1023.29s.

Key takeaway

For fMRI researchers developing population-level decoders, SpectralOT offers a robust and efficient solution. Its 30.5x faster computation than FUGW and easier alpha parameter tuning streamline experimental iterations. You can achieve superior cross-subject decoding and inter-subject correlation by balancing functional and geometric constraints, making it ideal for smaller-scale studies and broader applications. Consider its integration into deep learning workflows for future advancements.

Key insights

SpectralOT leverages cortical geometry via Laplace-Beltrami eigenmodes for fast, accurate, and tunable fMRI functional alignment.

Principles

• Functional architecture is not strictly tied to anatomical organization.
• Combining anatomical and functional constraints improves alignment quality.
• Entropic regularization can increase inter-subject correlation.

Method

SpectralOT computes geometric and functional dissimilarity matrices, interpolates them with a single parameter alpha, then uses a Sinkhorn Entropic Optimal Transport solver to derive a coupling matrix.

In practice

• Apply SpectralOT for faster fMRI functional alignment.
• Tune the alpha parameter (0-1) for balanced functional and anatomical constraints.
• Integrate SpectralOT into deep learning frameworks due to differentiability.

Topics

• fMRI
• Functional Alignment
• Optimal Transport
• Laplace-Beltrami Operator
• Brain Decoding
• Cortical Geometry

Code references

• nmwsharp/robust-laplacians-py
• pbarbarant/spectralot

Best for: AI Scientist, Research Scientist, Machine Learning Engineer

Related on AIssential

• Seeing the imagined: a latent functional alignment in visual imagery decoding from fMRI data — Latent functional alignment significantly improves fMRI-based visual imagery decoding by leveraging perception-learned semantic structures.
• fMRI2Face: A Full-HD fMRI-Video Dataset and Geometry-Guided Neural Decoding Framework for Dynamic Human Face Reconstruction — A new fMRI dataset and geometry-guided neural framework enable high-fidelity reconstruction of dynamic human faces from brain activity.
• SUP-MCRL: Subject-aware Unified Pseudo-feature Coded Multimodal Contrastive Representation Learning for EEG Visual Decoding — SUP-MCRL enhances EEG visual decoding by integrating semantic, subject-adaptive, and collapse-prevention mechanisms for robust multimodal representation learning.
• Neural Visual Decoding via Cognitive guided Adaptive Blurring and Information Constrained Alignment — CAIA enhances EEG-based visual decoding by adaptively blurring images and filtering EEG signals based on cognitive principles.
• Harnessing cortical geometry, wiring, and function as inductive biases for recurrent neural networks — Cortical geometry, wiring, and function serve as powerful inductive biases for more effective recurrent neural networks.
• Nonlocal operator learning for fMRI encoding and decoding tasks — Nonlocal neural integral operators effectively model fMRI dynamics, benefiting from broader spatiotemporal context for improved prediction and representation.

See Counsel's argued verdicts on the open AI decisions leaders are weighing →

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by stat.ML updates on arXiv.org.