Fast level-set based image segmentation using coherent propagation

Abstract

A generalized coherent propagation algorithm for level set evolution yielded substantial improvement in processing time with both synthetic datasets and medical images.

Multiorgan Segmentation Using Coherent Propagating Level Set Method Guided by Hierarchical Shape Priors and Local Phase Information

“…However, the processing time is still in the range of 10-30 min to segment a single organ from a voxel image [22]. To speed up level set segmentation, we proposed a fast level set method using coherent propagation [20,22]. The new method not only limits the computation to a narrow band, but also eliminates the points that have reached the object border.…”

“…Within a single period, once W (x) returns 0, at point x, then this point is excluded from the further computation, until at least one of its neighbours' level set value has changed. The ability to detect the converged points helps the new level set method to achieve at least 10 times speedup in various segmentation tasks when compared with the sparse field level set algorithm [22].…”

“…The most popular implementation of this type is the sparse field level set, which only updates the level set function on a 1-pixel wide band that the zero level set passes through [24]. However, the processing time is still in the range of 10-30 min to segment a single organ from a voxel image [22]. To speed up level set segmentation, we proposed a fast level set method using coherent propagation [20,22].…”

“…Within a single period, its value is fixed to 1 or −1. The initial value of Tr can be either estimated using the average external speed in a neighbourhood [22] or assigned via user interaction, such as dragging the contour inwards or outwards [23]. In the latter case, the trend of the whole contour is synchronized, meaning that the contour is designed to only expand or shrink in the first period.…”

“…The new algorithm forces the contour to move monotonically according to a predicted developing trend which makes the level set functions converge faster. It also makes it possible to detect local convergence, so that the parts of the boundary that have reached their final position can be excluded in subsequent iterations, thus significantly reducing computation time [20,22]. The proposed method was tested using the VISCERAL benchmark database, and promising results were delivered within reasonable processing time without any user intervention.…”