Search Papers | Poster Sessions | All Posters
Poster A28 in Poster Session A - Tuesday, August 6, 2024, 4:15 – 6:15 pm, Johnson Ice Rink
Bayesian inference for correlations between brain activity patters
Jörn Diedrichsen1 (), Mahdiyar Shahbazi1; 1Western Institute of Neuroscience, Western University, London, Ontario.
The relationship between activity patterns in response to different conditions provides important insights into the computations occurring in a brain area. To fully characterize the representational geometry, it is often desirable to establish the correlation between two activity patterns (or between two representational hyper-planes), independent of the size of the activation. Traditional point-estimates of correlation coefficients between patterns are biased and not suited for inference. This is especially true for functional magnetic resonance imaging (fMRI) data, which is corrupted by substantial measurement noise. Here we propose a Bayesian approach, which approximates the posterior distribution of the correlation coefficient. This approach allows valid inferences, both when comparing a correlation coefficient against a fixed value (one-sample problem), as well as comparing two correlation coefficients across two different regions or groups of subjects (two-sample problem). The utility of the approach is demonstrated through the reanalysis of a number published imaging studies.
Keywords: multivariate analysis representational similarity analysis fMRI Bayesian inference