Estimating fMRI Timescale Maps.

Area of Science:

• Neuroscience
• Computational Neuroscience
• Data Science

Background:

• Brain activity occurs over hierarchical timescales, crucial for information processing and linking brain organization.
• Current methods for estimating these timescales in fMRI data rely on restrictive assumptions and lack robust statistical inference capabilities.

Purpose of the Study:

• To formalize and evaluate two novel methods for mapping brain timescales in resting-state fMRI data.
• To overcome limitations of existing methods by relaxing assumptions and incorporating standard errors for statistical inference.

Main Methods:

• Developed and evaluated two methods: a time-domain fit of an autoregressive (AR1) model and an autocorrelation-domain fit of an exponential decay model.
• Defined timescales by projecting fMRI time series onto approximating models, requiring only stationarity and mixing conditions.
• Incorporated robust standard errors to address model misspecification and enable statistical inference.

Main Results:

• The time-domain method demonstrated more accurate timescale estimates, especially under model misspecification.
• This method proved computationally efficient for high-dimensional fMRI data and produced maps consistent with known functional brain organization.
• Validated parameter recovery in simulations and demonstrated applications to Human Connectome Project fMRI data.

Conclusions:

• The study successfully enables valid statistical inference on fMRI timescale maps.
• The proposed methods offer improved accuracy, computational efficiency, and statistical rigor for analyzing brain dynamics.
• Provided Python implementations for broader research community adoption.