Isogeometric analysis (IGA), due to its high efficiency and accuracy, can be replaced by the conventional finite element method in the topology optimizations (TOs). In this paper, we present an efficient isogeometric TO method based on the scheme of bi-directional evolutionary structural optimization (BESO). The structural response analysis is implemented by the IGA, and the detailed derivation is introduced. Based on the local support property of the non-uniform rational B-spline (NURBS), a NURBS filter is proposed to smooth the sensitivity numbers, of which NURBS space is different from that used in the IGA as long as the element spans keep consistency. Three benchmark numerical examples with different filters and mesh sizes are presented to validate the stability and mesh-dependency of the proposed isogeometric TO method.