Simultaneous EEG-fMRI has been vastly used to investigate functional networks of brain combining high spatial resolution (millimeter) of fMRI with high temporal resolution (millisecond) of EEG. However, to extract the most relevant information from the acquired data, it is necessary to develop analysis methods with less ad hoc assumptions. To this end, brain rhythms are often used which are the specific frequency-bands of EEG signal and assumed to represent diverse sub-second cognitive processes at different parts of the cortex. Furthermore, single-trial analysis of EEG is believed to show more realistic picture of ongoing and event-related activities of brain. Here, we present a nonparametric multiple change-point detection and estimation method for the single-trial analysis of simultaneous EEG-fMRI experiment recorded during auditory and visual oddball tasks. In a simple attention task like oddball, the frontal cortex of brain is responsible to distinguish and respond appropriately to target versus standard events. By using EEG signal at the frontal cortex, we show that the α-band activity changes according to ″inhibition timing″ hypothesis and the β-band activity is in line with ″maintaining the status quo″ hypothesis. Furthermore, using these activities to build regressors in the GLM analysis of fMRI, we localize active brain regions with high spatial and temporal resolutions and elaborate further on the coordination of attentional networks across brain.