Primal/Dual Linear Programming and Statistical Atlases for Cartilage Segmentation

Abstract

Abstract. In this paper we propose a novel approach for automatic segmentation of cartilage using a statistical atlas and efficient primal/dual linear programming. To this end, a novel statistical atlas construction is considered from registered training examples. Segmentation is then solved through registration which aims at deforming the atlas such that the conditional posterior of the learned (atlas) density is maximized with respect to the image. Such a task is reformulated using a discrete set of deformations and segmentation becomes equivalent to finding the set of local deformations which optimally match the model to the image. We evaluate our method on 56 MRI data sets (28 used for the model and 28 used for evaluation) and obtain a fully automatic segmentation of patella cartilage volume with an overlap ratio of 0.84 with a sensitivity and specificity of 94.06% and 99.92%, respectively.

Shape-based acetabular cartilage segmentation: application to CT and MRI datasets

“…Up to now, many efforts have been made for cartilage segmentation such as the methods proposed in [6][7][8][9][10][11][12][13][14][15][16][17]. Some of these methods require a training set to get the prior knowledge about the shape, position, or texture of the cartilage [6][7][8][9][10][11][12][13], while others are based on the knowledge-free methods [14][15][16][17].…”

“…In [9], a segmentation scheme was proposed for knee cartilages from MR database that involves automatic segmentation of the bones using a three-dimensional (3D) ASM, extraction of the expected bone-cartilage interface (BCI), and cartilage segmentation using a deformable model. In [10], patella cartilage in MRI datasets was localized using a statistical atlas and efficient primal/dual linear programming. In [11], a fully automatic method was employed to segment MR knee cartilages based on a two-step k-nearest neighbor (k-NN) voxel classifier.…”

“…Some of these methods require a training set to get the prior knowledge about the shape, position, or texture of the cartilage [6][7][8][9][10][11][12][13], while others are based on the knowledge-free methods [14][15][16][17]. In [6], a multi-template-based label fusion method was used to automatically segment 15 hip cartilage tissues taken by the magnetic resonance imaging (MRI) technique.…”

“…In [17], a semiautomatic method based on the Bezier splines and the Canny edge detector was utilized to segment the knee cartilages from MR database. Knowledge-free methods have low accuracy level in the presence of noise and loss of information [14][15][16][17], while training-based methods are more accurate [6][7][8][9][10][11][12][13]. However, training-based methods have at least one of these limitations: (1) high user interaction needs [8], (2) effectiveness only for healthy cartilage datasets [8,9], (3) dependency on modality [6,7,[9][10][11][12][13], and (4) lack of attention to spatial relationship between the neighboring structures in order to increase the accuracy level and generalization ability [6,8,10,11].…”

Joint Model-Pixel Segmentation with Pose-Invariant Deformable Graph-Priors

“…The former class of methods models geometric variation of the object of interest and then seeks an instance of this space in the image. Active shape/appearance models [6,5] and atlas-based methods [7] are popular examples. Manifold enhanced methods aim to minimize the distance of the solution from the learned manifold, e.g.…”