Texture analysis plays an important role in many image processing tasks. In this work, we present a texture descriptor based on the topology of excursion sets, derived from the concept of Minkowski functionals, and evaluate their usefulness in the detection of breast masses in 2D breast ultrasound images. The application includes three major stages: preprocessing, including candidate generation through computation of gradient concentration under a Fisher-Tippet noise model (in itself another contribution of the paper); texture feature extraction; and region classification using a Random Forests classifier. Performance of the proposed method is evaluated on 135 2D BUS images with 139 masses. Our method reaches 91% sensitivity with an averaged 1.19 false detections, and the proposed texture feature compares favorably against the often-used grey level co-occurrence matrices on the exact the same task.

Breast cancer is one of the leading cause of cancer related deaths in women and early detection is crucial for reducing mortality rates. In this paper, we present a novel and fully automated approach based on tissue transition analysis for lesion detection in breast ultrasound images. Every candidate pixel is classified as belonging to the lesion boundary, lesion interior or normal tissue based on its descriptor value. The tissue transitions are modeled using a Markov chain to estimate the likelihood of a candidate lesion region. Experimental evaluation on a clinical dataset of 135 images show that the proposed approach can achieve high sensitivity (95 %) with modest (3) false positives per image. The approach achieves very similar results (94 % for 3 false positives) on a completely different clinical dataset of 159 images without retraining, highlighting the robustness of the approach.

Minkowski Functionals (MFs) are geometric measurements of 3D shapes, including volume, surface area, curvature and Euler number. MFs can be used as texture descriptors for medical image analysis in the segmentation of normal anatomy as well as in the detection/diagnosis of pathology. In this paper, we propose a method for fast computation of MFs based on integral images, which offers significantly improved accuracy and efficiency compared with previous works. In addition, MFs computed using our method are used in applications on image segmentation and pathology detection. Our experiment results clearly demonstrate the potential of MFs in such medical image analysis tasks.

This paper presents registration via embedded maps (REM), a deformable registration algorithm for images with varying topology. The algorithm represents 3-D images as 4-D manifolds in a Riemannian space (referred to as embedded maps). Registration is performed as a surface evolution matching one embedded map to another using a diffusion process. The approach differs from those existing in that it takes an a priori estimation of image regions where topological changes are present, for example lesions, and generates a dense vector field representing both the shape and intensity changes necessary to match the images. The algorithm outputs both a diffeomorphic deformation field and an intensity displacement which corrects the intensity difference caused by topological changes. Multiple sets of experiments are conducted on magnetic resonance imaging (MRI) with lesions from OASIS and ADNI datasets. These images are registered to either a brain template or images of healthy individuals. An exemplar case registering a template to an MRI with tumor is also given. The resulting deformation fields were compared with those obtained using diffeomorphic demons, where topological changes are not modeled. These sets of experiments demonstrate the efficacy of our proposed REM method for registration of brain MRI with severe topological differences.

In this paper, we develop a new deformable registration algorithm for images with pathology-induced topological changes. In this algorithm, 3D images are embedded as 4D surfaces in a Riemannian space and the registration is conducted as a surface evolution process. Our algorithm differs from existing methods in the sense that it takes an a-priori estimation of areas with topological change as an additional input and generates dense deformation vector fields which are free of false deformation. In particular, the output of our algorithm is composed of a diffeomorphic deformation field and an intensity displacement which corrects the intensity difference caused by the topological changes. The experiments demonstrate that our proposed algorithm is capable of accurately registering images with considerable topological changes. More importantly, the resulting deformation field is not impacted by topological changes, i.e., there is no false deformation.

The recent driven equilibrium single-pulse observation of T1 (DESPOT1) approach permits real-time clinical acquisition of large-volume and high-isotropic-resolution T1 mapping of MR tissue parameters with improved uniformity. It is assumed that the quantitative nature of maps will facilitate clinical applications such as disease diagnosis and comparison across subjects. However, there is not yet enough quantitative evidence on the actual benefit of adopting T1 maps, especially in computer-aided medical image analysis tasks. In this study, we compare methods with respect to image types, T1-weighted images or T1 maps, in automatic brain MRI segmentation. Our experimental results demonstrate that, using T1 maps, different segmentation algorithms show better agreement with each other, compared to that from using T1-weighted images. Furthermore, through multi-dimensional-scaling projection, we are able to visualize the relative affinity among segmentation results, which reveals that the projections of those segmentations using two different types of input images tend to form two separate clusters. Finally, by comparing to expert segmented reference segmentation of brain sub-regions, our results clearly indicate a better agreement between the manual reference and those automatic ones on T1 maps. In other words, our study provides an evidence for the hypothesis that compared to the conventionally used T1-weighted images, T1 maps lead to improved reliability in automatic brain MRI segmentation task.

Topological changes are common in brain MR images for aging or disease studies. For deformable registration algorithms, which are formulated as a variational problem and solved by the minimization of certain energy functional, topological changes can cause false deformation in the resulting vector field, and affect algorithm convergence. In this work, we focus on the effect of topological changes on diffeomorphic and inverse-consistent deformable registration algorithms, specifically, diffeomorphic demons and symmetric LDDMM. We first use a simple example to demonstrate the adverse effect of topological changes on these algorithms. Then, we propose an novel framework that can be imposed onto any existing diffeomorphic and inverse-consistent deformable registration algorithm. Our framework renders these registration algorithms robust to topological changes, where the outputwill consist of two components. The first is a deformation field that presents only the brain structural change which is the expected vector field if the topological change did not exist. The second component is a label map that provides a segmentation of the topological changes appeared in input images.

We present a face recognition method based on sparse representation for recognizing 3D face meshes under expressions using low-level geometric features. First, to enable the application of the sparse representation framework, we develop a uniform remeshing scheme to establish a consistent sampling pattern across 3D faces. To handle facial expressions, we design a feature pooling and ranking scheme to collect various types of low-level geometric features and rank them according to their sensitivities to facial expressions. By simply applying the sparse representation framework to the collected low-level features, our proposed method already achieves satisfactory recognition rates, which demonstrates the efficacy of the framework for 3D face recognition. To further improve results in the presence of severe facial expressions, we show that by choosing higher-ranked, i.e., expression-insensitive, features, the recognition rates approach those for neutral faces, without requiring an extensive set of reference faces for each individual to cover possible variations caused by expressions as proposed in previous work. We apply our face recognition method to the GavabDB and FRGC 2.0 databases and demonstrate encouraging results.

Abstract The performances of most existing face recognition systems suffer from facial expressions. Unfortunately, there is not yet a satisfactory solution. Therefore, the main focus of this thesis is on expression-invariant face recognition algorithms. In this thesis, we first propose a 2D face recognition algorithm by separately modeling geometry and texture information in a face image. The effect of expression is removed from