Efficient level set formulation for segmentation and correction with application to medical images

摘要

Accurate medical image segmentation can greatly improve doctors' speed of diagnosis and diagnosis rate. But the medical image is usually accompanied by the intensity inhomogeneity, which seriously interferes with the accuracy of segmentation results. In this paper, we propose a novel active contour model with the level set formulation to deal with this problem. With the bias field added into the energy functional, our model not only can accurately segment inhomogeneous images, but also can effectively eliminate the intensity inhomogeneity to get homogeneous correction images. Since our energy functional has a special form similar to the L1 regularization problem, we prefer to apply the split Bregman method to efficiently minimize the energy functional. Then, we use a variety of medical images to test the performance of our model. Experimental results demonstrate that our model can be applied in medical images with satisfactory results. Besides, qualitative and quantitative comparisons with the LSE model further demonstrate the superiority of our model in segmentation accuracy, correction effect and efficiency. The robustness to initial contour and noises is also verified to be the outstanding advantage of our model.