Aiming at the characteristics of the intensity inhomogeneous and diversiform parathyroid lesions in the ultrasound images of the parathyroid gland, we propose a hybrid level set model for parathyroid gland segmentation based on local entropy of images. The proposed model uses both global and local image information. To address the problem of the inhomogeneous intensity distribution in ultrasound images, local entropy of images is used to determine the weight of the global term to improve the model's adaptivity. In addition, two scales arc adopted to prevent over-segmentation and calculation inefficiency on the large and small scales, respectively. Experimental results show that the proposed model can adapt to different ultrasound images of parathyroid gland, which makes the evolution curve converge to the target contour automatically. In addition, this model has high segmentation accuracy and computational efficiency.