The ability to switch between different behaviors is essential to all animals’ survival. The selection of a behavior is guided by multiple factors, such as sensory inputs, internal states, and memory, which suggests that regions across the brain are involved in the decision to switch between them. While whole-brain information is necessary to investigate the neural basis of behavioral switching, brain-wide imaging in behaving mice has proved challenging so far. We employed functional Ultrasound imaging (fUS) to record large-scale neural dynamics in head-fixed mice while simultaneously tracking their behavioral state. Our aim is to identify brain regions that predict self-initiated behavioral transitions that occur in the absence of external triggers. Accordingly, we utilized the virtual burrow assay in which head-fixed mice are placed in an air-floating tube, from which they can voluntarily egress. Leveraging the behavior segmentation framework VAME, we found that mice (N = 11, 60 sessions) robustly exhibit distinct behavioral states in this assay that include egress, whisking, inactivity, and grooming. Utilizing brain-wide fUS, we subsequently observed activity patterns associated with these distinct behavioral states and performed whole-brain time-resolved decoding around behavioral transitions. Our results show that whole-brain activity can predict an uninstructed switch in behavior several seconds before its onset, indicating that a change in behavior is preceded by a detectable change in brain state. Furthermore, region-wise decoding and fiberphotometry revealed specific brain areas that best predict behavioral transitions. Through this unbiased approach, our work sheds light on the neural dynamics preceding uninstructed transitions of behavioral state.