Robust Adaptive Segmentation of 3D Medical Images with Level Sets

This paper is concerned with the use of the Level Set formalism to segment anatomical structures in 3D medical images (ultrasound or magnetic resonance images). A closed 3D surface propagates towards the desired boundaries through the iterative evolution of a 4D implicit function. The major contribution of this work is the design of a robust evolution model based on adaptive parameters depending on the data. First the step size and the external propagation force factor, both usually predetermined constants, are automatically computed at each iteration. Additionally, regionbased information, rather than spatial image gradient, is exploited by estimating intensity probability density functions over the image. As a result, the method can be applied to various kinds of data. Quantitative and qualitative results on brain MR images and 3D echographies of carotid arteries are reported and discussed. Key-words: 3D segmentation, deformable models, level sets, intensity distribution, brain MRI, 3D ultrasound images