Principal spatiotemporal pattern mapping detects brain activity partially decoupled with task timing

Neural activity arises from interactions between bottom-up sensory inputs and top-down information process, and thus may not be strictly time-locked to external task stimuli in function MRI (fMRI). Such temporal decoupling between stimuli and neural responses challenges the generalized linear model (GLM) that assumes fMRI signals as a convolution of a prior hemodynamic response function (HRF) with task timings, potentially leading to incomplete detections of task-evoked brain activations. To address this issue, we developed an HRF-model-free framework that detects task-evoked brain activations by capturing spatiotemporal pattern of fMRI signals within functional topography in the brain. Using this framework, we investigated neural dynamics of participants engaged in Human Connectome Project (HCP) emotion tasks. We identified neural activations in prefrontal subregions that exhibit partial temporal decoupling from task timing, rendering them undetectable by the GLM. These activations are likely driven by implicit events emerging from top-down information processes following the reception of external stimuli. Together, these findings suggest that task-evoked neural dynamics are shaped not only by externally events but also by autonomous cognitive process, underscoring the complex and interactive nature of brain function and highlighting the need for new analytical frameworks for task-based fMRI.