Phantom sound perception is the perception of a sound in the absence of a corresponding external sound source. It is a common symptom for which no treatment exists. Gaining a better understanding of its pathophysiology by applying network science might help in identifying targets in the brain for neuromodulatory approaches to treat this elusive symptom. Brain networks are commonly organized as functional modules which have a densely connected core network coupled to a communally-organized peripheral network. The core network is called the rich club network and the peripheral network is divided into the feeder and local networks. In current study, we investigate the effects of virtual lesions on the endogenous dynamics, complexity and robustness of the remaining brain. It is hypothesized that depending on whether nodes is functionally central to the network or not, the robustness and dynamics of the network change when a lesion in introduced. We therefore investigate the effect of introducing a virtual focal lesion randomly to different nodes is in the tinnitus network and contrast it to the effect of specifically targeting the nodes of the rich-club, feeder and local nodes in patients experiencing a phantom sound (i.e. tinnitus). The tinnitus and control networks were computed from the source-localized EEG of 311 tinnitus patients and 256 control subjects. The results of the current study indicate that both the tinnitus and control networks are robust to the attack on random and rich club nodes, but are drastically modified when attacked from the periphery, especially while targeting the feeder hubs. In both the tinnitus and control networks, feeder nodes were found to have a higher betweenness centrality value than the rich club nodes. This shows that the feeders have a larger influence on the information transmission through the brain than the rich club nodes, by transferring information from the peripheral communities to the core. Further, evidence for the theoretical model of a multimodal tinnitus network is also presented showing that the tinnitus network is divided into individual, separable modules each possibly encoding a different aspect of tinnitus. The current study alludes to the concept that the efficient modification of the tinnitus network is theoretically possible by disconnecting the individual communities from the core of the pathological network.