Robust global minimization of active contour model for multi-object medical image segmentation

The performance of active contour model is limited to the zero level set functions and energy weight coefficients, which affect the accuracy of the segmentation results seriously. A new robust global minimization of active contour model is proposed in this paper to eliminate those limitations for multi-object medical image segmentation. The zero level set functions are initialized with the coarse results extracted by spatial fuzzy C-means clustering, while the energy weight coefficients are also estimated with the coarse results. The level set evolution starts from the regions near the true boundaries. Therefore, the energy functional converges to global minimum instead of falling into local minimum. These improvements lead to more robust segmentation results. The proposed algorithm is verified with multi-object medical image segmentation. The results demonstrate that compared with traditional algorithm the performance of the proposed algorithm is better for multi-object medical image segmentation in presence of complex shapes and weak boundaries.