Shortest-Path Constraints for 3D Multiobject Semiautomatic Segmentation Via Clustering and Graph Cut

Abstract

To cite this version:Razmig Kéchichian, Sébastien Valette, Michel Desvignes, Rémy Prost. Shortest-path constraints for 3D multiobject semiautomatic segmentation via clustering and Graph Cut. IEEE Transactions on Image Processing, Institute of Electrical and Electronics Engineers, 2013, 22 (11) Abstract-We derive shortest-path constraints from graph models of structure adjacency relations and introduce them in a joint centroidal Voronoi image clustering and Graph Cut multiobject semiautomatic segmentation framework. The vicinity prior model thus defined is a piecewise-constant model incurring multiple levels of penalization capturing the spatial configuration of structures in multiobject segmentation. Qualitative and quantitative analysis and comparison with a Potts prior-based approach and our previous contribution on synthetic, simulated and real medical images show that the vicinity prior allows for the correct segmentation of distinct structures having identical intensity profiles and improves the precision of segmentation boundary placement while being fairly robust to clustering resolution. The clustering approach we take to simplify images prior to segmentation strikes a good balance between boundary adaptivity and cluster compactness criteria furthermore allowing to control the trade-off. Compared to a direct application of segmentation on voxels, the clustering step improves the overall runtime and memory footprint of the segmentation process up to an order of magnitude without compromising the quality of the result.

Automatic Multiorgan Segmentation Using Hierarchically Registered Probabilistic Atlases

“…Therefore, we simplify the image prior to segmentation by an image-adaptive centroidal Voronoi tessellation (CVT), which strikes a good balance between cluster compactness and object boundary adherence and helps to place subsequent segmentation boundaries precisely. We have shown that the clustering step improves the overall run-time and memory footprint of the segmentation process up to an order of magnitude without compromising the quality of the result [14].…”

“…11.3 encode prior information on interactions between labels assigned to pairs of neighbouring variables encouraging the spatial consistency of labelling with respect to a reference model. We define these terms according to the piecewise-constant vicinity prior model proposed in [14], which, unlike the standard Potts model, incurs multiple levels of penalization capturing the spatial configuration of structures in multiobject segmentation. It is defined as follows.…”

“…Traditionally, single-object-or pathology-oriented, recent image processing methods [9,12,14,15,19,23,25,27] have made the analysis and the segmentation of multiple anatomical structures increasingly possible. However, CT and MR images have intrinsic characteristics that render their automatic segmentation challenging.…”

“…It is impossible to identify and segment such structures automatically on the basis of intensity information only. Hence, most advanced segmentation methods exploit some form of prior information on structure location [12,19,27] or interrelations [9,14,23,25] to achieve greater robustness and precision. Hierarchical approaches to segmentation [23,25,32] rely on hierarchical organizations of prior information and algorithms that proceed in a coarse-to-fine manner according to anatomical level of detail.…”

“…The spatial prior is derived from shortest-path pairwise constraints defined on the adjacency graph of structures [14], and the organ location likelihood is defined by probabilistic atlases [24] learned from the VISCERAL training dataset [11]. We register organ atlases to the image prior to segmentation using a fast (2+1)D registration method based on SURF keypoints.…”