Medical Image Segmentation Using Magnetostatic Active Contours (MAC) with Tensor Diffusion

In medical imagery, traditional deformable models often face substantial challenges due to fine structures and image complexity. Recently, based on magnetostatic theory, a new deformable model, namely MAC, is proposed for improving the ability of the active contour in dealing with complex geometries and segmentation difficulties. A Laplacian diffusion scheme is proposed in the MAC model to tackle excessive image noise which can interrupt image gradient vectors and in turn affect the external force field. In this paper, a derived vector potential field (VPF) is employed to obtain magnetic force and thus a diffusion tensor can be applied to diffuse VPF in terms of both magnitude and directional information, instead of directly diffusing the magnetic field as in the MAC model. Our diffusion is carried out both in spatial and temporal aspects of VPF so that the performance of the deformable model is significantly improved while images are with low signal-noise ratio (SNR) and poor contrast. In addition, the proposed diffusion enhancement can lead to evolving the curve smoothly and thus level set evolution is adapted to approach genuine object of interest. By applying in several medical image modalities, the results demonstrate the effectiveness of the proposed method.