As the Distance Regularized Level Set Evolution (DRLSE) model is difficult to deal with weak boundary images and has inefficient curve evolution, a novel ultrasound image segmentation model based on the phase information in a variational level set formulation is proposed. The proposed model constructs a new boundary indicator function by using the theory of phase congruency detection, instead of the boundary stopping function in the model of DRLSE, and then come to the new energy functional. Experimental results show that the method accelerates the evolution speed. Furthermore, it can extract thyroid tumors ideally in the segmentation of ultrasound images.