An empirical evaluation of functional alignment using inter-subject decoding.

Area of Science:

• Neuroscience
• Computational Neuroscience
• Brain Imaging Analysis

Background:

• Inter-individual variability in brain functional organization hinders the discovery of generalizable neural coding principles.
• Functional alignment methods, which match neural signals based on functional similarity, offer a promising solution to this challenge.
• The comparative performance of existing functional alignment techniques remains unclear.

Purpose of the Study:

• To benchmark five functional alignment methods for inter-subject decoding accuracy.
• To introduce and evaluate two novel functional alignment extensions: piecewise Shared Response Modelling (SRM) and intra-subject alignment.
• To assess the computational efficiency and scalability of the evaluated methods.

Main Methods:

• Benchmarking of three existing methods: piecewise Procrustes, searchlight Procrustes, and piecewise Optimal Transport.
• Introduction and benchmarking of two new methods: piecewise Shared Response Modelling (SRM) and intra-subject alignment.
• Evaluation on four publicly available neuroimaging datasets, assessing inter-subject decoding accuracy and computational performance.

Main Results:

• Functional alignment generally enhances inter-subject decoding accuracy.
• Piecewise Shared Response Modelling (SRM) and Optimal Transport demonstrated strong performance at both region-of-interest and whole-brain levels.
• Method performance varied depending on the specific research context, highlighting the need for careful selection.

Conclusions:

• Functional alignment is a valuable strategy for improving inter-subject comparisons in neuroscience.
• SRM and Optimal Transport are effective methods for aligning functional brain data across individuals.
• The study provides open implementations, facilitating the use of these methods in future research.