Learned Shape Models Research Articles

Fast 3D Liver Segmentation Using a Trained Deep Chan-Vese Model

This paper introduces an approach for 3D organ segmentation that generalizes in multiple ways the Chan-Vese level set method. Chan-Vese is a segmentation method that simultaneously evolves a level set while fitting locally constant intensity models for the interior and exterior regions. First, its simple length-based regularization is replaced with a learned shape model based on a Fully Convolutional Network (FCN). We show how to train the FCN and introduce data augmentation methods to avoid overfitting. Second, two 3D variants of the method are introduced, one based on a 3D U-Net that makes global shape modifications and one based on a 3D FCN that makes local refinements. These two variants are integrated in a full 3D organ segmentation approach that is capable and efficient in dealing with the large size of the 3D volumes with minimal overfitting. Experiments on liver segmentation on a standard benchmark dataset show that the method obtains 3D segmentation results competitive with the state of the art while being very fast and having a small number of trainable parameters.

Affine-invariant Shape Recognition Using Grassmann Manifold

Traditional Kendall shape space theory is only applied to similar transform. However, geometric transforms of the object in the imaging process should be represented by affine transform at most situations. We analyze the nonlinear geometry structure of the affine invariant shape space and propose an affine-invariant shape recognition algorithm based on Grassmann manifold geometry. Firstly, we compute the mean shape and covariance for every shape class in the train sets. Then, we construct their norm probability models on the tangent space at each mean shape. Finally, we compute the maximum likelihood class according to the measured object and prior learned shape models. We use the proposed algorithm to recognize shapes in standard shape dataset and real images. Experiment results on MPEG-7 shape dataset show that our recognition algorithm outperforms the algorithm based on Procrustean metric in traditional Kendall shape space theory. Experiment results on real images also show that the proposed algorithm exhibits higher capacity to affine transform than the Procrustean metric based algorithm and can recognize object classes with higher posterior probability.

A hybrid blob-slice model for accurate and efficient detection of fluorescence labeled nuclei in 3D

BackgroundTo exploit the flood of data from advances in high throughput imaging of optically sectioned nuclei, image analysis methods need to correctly detect thousands of nuclei, ideally in real time. Variability in nuclear appearance and undersampled volumetric data make this a challenge.ResultsWe present a novel 3D nuclear identification method, which subdivides the problem, first segmenting nuclear slices within each 2D image plane, then using a shape model to assemble these slices into 3D nuclei. This hybrid 2D/3D approach allows accurate accounting for nuclear shape but exploits the clear 2D nuclear boundaries that are present in sectional slices to avoid the computational burden of fitting a complex shape model to volume data. When tested over C. elegans, Drosophila, zebrafish and mouse data, our method yielded 0 to 3.7% error, up to six times more accurate as well as being 30 times faster than published performances. We demonstrate our method's potential by reconstructing the morphogenesis of the C. elegans pharynx. This is an important and much studied developmental process that could not previously be followed at this single cell level of detail.ConclusionsBecause our approach is specialized for the characteristics of optically sectioned nuclear images, it can achieve superior accuracy in significantly less time than other approaches. Both of these characteristics are necessary for practical analysis of overwhelmingly large data sets where processing must be scalable to hundreds of thousands of cells and where the time cost of manual error correction makes it impossible to use data with high error rates. Our approach is fast, accurate, available as open source software and its learned shape model is easy to retrain. As our pharynx development example shows, these characteristics make single cell analysis relatively easy and will enable novel experimental methods utilizing complex data sets.

Embedding Gestalt laws in Markov random fields

The goal of this paper is to study a mathematical framework of 2D object shape modeling and learning for middle level vision problems, such as image segmentation and perceptual organization. For this purpose, we pursue generic shape models which characterize the most common features of 2D object shapes. In this paper, shape models are learned from observed natural shapes based on a minimax entropy learning theory. The learned shape models are Gibbs distributions defined on Markov random fields (MRFs). The neighborhood structures of these MRFs correspond to Gestalt laws-colinearity, cocircularity, proximity, parallelism, and symmetry. Thus, both contour-based and region-based features are accounted for. Stochastic Markov chain Monte Carlo (MCMC) algorithms are proposed for learning and model verification. Furthermore, this paper provides a quantitative measure for the so-called nonaccidental statistics and, thus, justifies some empirical observations of Gestalt psychology by information theory. Our experiments also demonstrate that global shape properties can arise from interactions of local features.