In independent component analysis (ICA)

a statistical latent variables model is employed to assume that the obtained data is a linear mixture of signals. To deal with the detection error when the nonlinear mixing signals of functional magnetic resonance imaging (fMRI) data are decomposed by means of ICA

a novel ICA method based on the instantaneous power of fMRI data is developed. Firstly

fMRI data are converted into its energy signals according to the energy form of electricity. Secondly

according to the relationship between blood oxygenation level dependent (BOLD) and T*2 signal

two types of instantaneous power of fMRI signals which represent the energy fluctuations of BOLD are proposed. Finally

based on the instantaneous power of fMRI data

the components correlated with the energies of brain activations are obtained by using a spatial ICA method. The effectiveness and advantage are elucidated through theoretical analyses and simulation tests

and it is also applied to vivo experimental epileptic fMRI

the results show that our method can robustly detect abnormal brain activities at resting state compared with the traditional ICA methods.