Temporal synchronization analysis: a model-free method for detecting robust and nonlinear brain activation in fMRI data

Abstract

The sluggishness of the fMRI blood oxygenation level dependent (BOLD) signal has motivated the use of block or trial-based experimental designs that rely on the assumption of linearity, typically modeled using the General Linear Model (GLM). But many non-sensory brain regions and subcortical areas do not correspond to such linearities. We introduce a model-free estimation method called Temporal Synchronization Analysis (TSA) which detects significant brain activations across trials and subjects at an individual time point. We validate it across multiple cognitive tasks (combined n=1600). In constrained task stimuli like visual checkerboard paradigms, we discovered novel nonlinearities not reported previously. In model-free task paradigms like listening to naturalistic auditory stimuli, TSA can detect unique stimuli linked quasi-temporal activations across default mode and language networks. Our user-friendly Python toolkit enables cognitive neuroscience researchers to identify stable and robust brain activation across various cognitive paradigms that are challenging to model with current methods.