Epilepsy is a neurological condition with a prevalence of 1%, and 14-34% have medically refractory epilepsy (MRE). Seizures in focal MRE are generated by a single epileptogenic zone (or focus), thus there is potentially a curative procedure - surgical resection. This procedure depends significantly on correct identification of the focus, which is often uncertain in clinical practice. In this study, we analyzed intracranial stereotaxic EEG (sEEG) data recorded in two human patients with drug-resistant epilepsy prior to undergoing resection surgery. We view the sEEG data as samples from the brain network and hypothesize that seizure foci can be identified based on their network connectivity during seizure. Specifically, we computed a time sequence of connectivity matrices from EEG recordings that represent network structure over time. For each patient, connectivity between electrodes was measured using the coherence in a given frequency band. Matrix structure was analyzed using singular value decomposition and the leading singular vector was used to estimate each electrode's time dependent centrality (importance to the network's connectivity). Our preliminary study suggests that seizure foci may be the most weakly connected regions in the brain during the beginning of a seizure and the most strongly connected regions towards the end of a seizure. Additionally, in one of the patients analyzed, the network connectivity under anesthesia highlights seizure foci. Ultimately, network centrality computed from sEEG activity may be used to develop an automated, reliable, and computationally efficient algorithm for identifying seizure foci.