This paper proposes a fast, accurate and automatic method to segment wood knots from images obtained by X-Ray Computed Tomography scanner. The wood knot segmentation is a classical problem where the most popular segmentation techniques produce unsatisfactory results. In a previous work, a method was developed to detect knot areas and an approach was proposed to segment the knots. However this last step is not entirely satisfactory in the presence of sapwood. This paper presents a novel approach for knot segmentation, based on the original idea considering slices tangential to the growth rings. They allow to track the knot from the log pith to the bark. Knots are then segmented by detecting discrete ellipses in each slice. A complete implementation is proposed on the TKDetection software available online.