Skip to Main Content
The segmentation of MR images is of great interests in automatic medical diagnosis. However, such images are corrupted by Rician noise and with fuzzy edges. The non-additive and intensity dependant features of Rician noise make image processistandardng very challenging. In this paper, a combination of techniques are carefully selected, tailored, and organized to improve the image quality for automatic segmentation of the articular cartilage from noisy MR images. A prototype software procedure is developed based on that. First, the Rician noise, known as the major factor contributing to corrupt the magnitude MR images, is investigated. A method based on the least squares approach is employed to estimate the noise standard deviation from the background mode of the image. Then, a total variation noise removal algorithm using iterative scheme is applied to remove the noise. After that, the vector field convolution active contour method is applied to the resultant image for cartilage segmentation. Two different approaches are proposed to define the initial contour to avoid local optimization traps. A test set MR images on 62 slices of human knee is used to illustrate the proposed system. Effectiveness of the proposed procedure is demonstrated using computational experiments in comparison to some existing methods.