Functional magnetic resonance imaging (fMRI) measures the brain activity by detecting signal changes based on the blood-oxygen level dependent (BOLD) contrast. However, this technique frequently involves both noise and actual neuronal activation when detecting real brain activity. Thus, identifying the underlying brain signal is critical. Many researches related to this initiative used different clustering techniques in whole time and spatial domain to detect the activated signals. However, the limitation of extending the clustering algorithm into whole domain is that the noise can significantly affect the cluster result. Moreover, many statistical methods would weaken either temporal or spatial solution of fMRI. The current research will propose a novel application of using time constrained clustering to reduce the noise influence by partitioning the functional MRI signals into connected time windows within which the signal information is essentially unchanged, allowing the analysis of real brain pattern in each static time window so that the method can consider both temporal and spatial effect while minimizing the effect of noise.