Search

Kernel k-means is an extension of the standard k-means clustering algorithm that identifies nonlinearly separable clusters. In order to overcome the duster initialization problem associated with this method, we propose the global kernel k-means algorithm, a deterministic and incremental approach to kernel-based clustering. Our method adds one cluster at each stage, through a global search procedure consisting of several executions of kernel k-means from suitable initializations. This algorithm does not depend on duster initialization, identifies nonlinearly separable dusters, and, due to its incremental nature and search procedure, locates near-optimal solutions avoiding poor local minima. Furthermore, two modifications are developed to reduce the computational cost that do not significantly affect the solution quality. The proposed methods are extended to handle weighted data points, which enables their application to graph partitioning. We experiment with several data sets and the proposed approach compares favorably to kernel k-means with random restarts. Index Terms--Clustering, graph partitioning, k-means, kernel k-means.

Sparse kernel methods are very efficient in solving regression and classification problems. The sparsity and performance of these methods depend on selecting an appropriate kernel function, which is typically achieved using a cross-validation procedure. In this paper, we propose an incremental method for supervised learning, which is similar to the relevance vector machine (RVM) but also learns the parameters of the kernels during model training. Specifically, we learn different parameter values for each kernel, resulting in a very flexible model. In order to avoid over-fitting, we use a sparsity enforcing prior that controls the effective number of parameters of the model. We present experimental results on artificial data to demonstrate the advantages of the proposed method and we provide a comparison with the typical RVM on several commonly used regression and classification data sets. Index Terms--Classification, kernel learning, regression, relevance vector machine (RVM), sparse Bayesian learning.

A new approach is proposed for image segmentation that is based on a hierarchical and spatially variant mixture model. The results have shown that this approach has improved not only finite mixture model (FMM) segmentation, but also its spatially variant version.

Two image restoration algorithms based on the Bayesian approach and a new hierarchical spatially adaptive image prior are proposed. Numerical experiments are conducted to compare the algorithms with other stationary and nonstationary algorithms, which demonstrate the efficiency of the proposed algorithms.

An ultrasound wearable system for remote monitoring and acceleration of the healing process in fractured long bones is presented. The so-called USBone system consists of a pair of ultrasound transducers, implanted into the fracture region, a wearable device and a centralized unit. The wearable device is responsible to carry out ultrasound measurements using the axial-transmission technique and initiate therapy sessions of low-intensity pulsed ultrasound. The acquired measurements and other data are wirelessly transferred from the patient-site to the centralized unit, which is located in a clinical setting. The evaluation of the system on an animal tibial osteotomy model is also presented. A dataset was constructed for monitoring purposes consisting of serial ultrasound measurements, follow-up radiographs, quantitative computed tomography-based densitometry and biomechanical data. The animal study demonstrated the ability of the system to collect ultrasound measurements in an effective and reliable fashion and participating orthopaedic surgeons accepted the system for future clinical application. Analysis of the acquired measurements showed that the pattern of evolution of the ultrasound velocity through healing bones over the postoperative period monitors a dynamic healing process. Furthermore, the ultrasound velocity of radiographically healed bones returns to 80% of the intact bone value, whereas the correlation coefficient of the velocity with the material and mechanical properties of the healing bone ranges from 0.699 to 0.814. The USBone system constitutes the first telemedicine system for the out-hospital management of patients sustained open fractures and treated with external fixation devices. Index Terms--Fracture healing, remote patient monitoring, ultrasound, wearable devices.

The Bayesian framework is used to address the blind image deconvolution (BID) problem for which three algorithms are presented to solve the proposed Bayesian problem in closed form that can be implemented in the discrete Fourier domain. Numerical experiments show that these algorithms yield promising improvements as compared to previous BID algorithms

We present probabilistic models which are suitable for class conditional density estimation and can be regarded as shared kernel models where sharing means that each kernel may contribute to the estimation of the conditional densities of all classes. We first propose a model that constitutes an adaptation of the classical radial basis function (RBF) network (with full sharing of kernels among classes) where the outputs represent class conditional densities. In the opposite direction is the approach of separate mixtures model where the density of each class is estimated using a separate mixture density (no sharing of kernels among classes). We present a general model that allows for the expression of intermediate cases where the degree of kernel sharing can be specified through an extra model parameter. This general model encompasses both above mentioned models as special cases. In all proposed models the training process is treated as a maximum likelihood problem and expectation-maximization (EM) algorithms have been derived for adjusting the model parameters. Index Terms--Classification, density estimation, expectation-maximization (EM) algorithm, mixture models, probabilistic neural networks, radial basis function (RBF) network.

Partial differential equations (PDEs) with boundary conditions (Dirichlet or Neumann) defined on boundaries with simple geometry have been successfully treated using sigmoidal multilayer perceptrons in previous works. This article deals with the case of complex boundary geometry, where the boundary is determined by a number of points that belong to it and are closely located, so as to offer a reasonable representation. Two networks are employed: a multilayer perceptron and a radial basis function network. The later is used to account for the exact satisfaction of the boundary conditions. The method has been successfully tested on two-dimensional and three-dimensional PDEs and has yielded accurate results. Index Terms--Boundary value problems, engineering problems, irregular boundaries, neural networks, partial differential equations (PDEs), penalty method, multilayer perceptron, radial basis function (RBF) networks.

Multiview clustering partitions a dataset into groups by simultaneously considering multiple representations (views) for the same instances. Hence, the information available in all views is exploited and this may substantially improve the clustering result obtained by using a single representation. Usually, in multiview algorithms all views are considered equally important, something that may lead to bad cluster assignments if a view is of poor quality. To deal with this problem, we propose a method that is built upon exemplar-based mixture models, called convex mixture models (CMMs). More specifically, we present a multiview clustering algorithm, based on training a weighted multiview CMM, that associates a weight with each view and learns these weights automatically. Our approach is computationally efficient and easy to implement, involving simple iterative computations. Experiments with several datasets confirm the advantages of assigning weights to the views and the superiority of our framework over single-view and unweighted multiview CMMs, as well as over another multiview algorithm which is based on kernel canonical correlation analysis. [ABSTRACT FROM AUTHOR]

In this paper, we present a new Bayesian model for the blind image deconvolution (BID) problem. The main novelty of this model is the use of a sparse kernel-based model for the point spread function (PSF) that allows estimation of both PSF shape and support. In the herein proposed approach, a robust model of the BID errors and an image prior that preserves edges of the reconstructed image are also used. Sparseness, robustness, and preservation of edges are achieved by using priors that are based on the Student's-t probability density function (PDF). This pdf, in addition to having heavy tails, is closely related to the Gaussian and, thus, yields tractable inference algorithms. The approximate variational inference methodology is used to solve the corresponding Bayesian model. Numerical experiments are presented that compare this BID methodology to previous ones using both simulated and real data. [ABSTRACT FROM AUTHOR]

An information brokerage environment for effective information structuring, indexing, and retrieval in the health-care administration sector is presented. The system is based on ontology modeling, natural language processing, extensible markup language, semantics analysis, and behavioral description. Semantics-based information acquisition is achieved through the uniform modeling, representation, and handling of domain-specific knowledge, both content-based and procedural. The system has been validated using information located on several repositories in the web and its performance is reported in terms of precision and recall. [ABSTRACT FROM AUTHOR]

Recent advents in magnetic resonance spectroscopy (MRS) techniques permit subsequent microarray analysis over the entire human transcriptome in the same tissue biopsies. However, extracting information from such immense quantities of data is limited by difficulties in recognizing and evaluating the relevant patterns of apparent gene expression in the context of the existing knowledge of phenotypes by histopathology. Using a quantitative approach derived from a knowledge base of pathology findings, we present a novel methodology used to process genome-wide transcription and MRS data. This methodology was tested to examine metabolite and genome-wide profiles in MRS and RNA in 55 biopsies from human subjects with brain tumors with ~100% certainty. With the guidance of histopathology and clinical outcome, 15 genes with the assistance of 15 MRS metabolites were able to be distinguished by tumor categories and the prediction of survival was better than when either method was used alone. This new method, combining MRS, genomics, statistics and biological content, improves the typing and understanding of the complexity of human brain tumors, and assists in the search for novel tumor biomarkers. It is an important step for novel drug development, it generates testable hypotheses regarding neoplasia and promises to guide human brain tumor therapy provided improved in vivo methods for monitoring response to therapy are developed.

Advancements in the diagnosis and prognosis of brain tumor patients, and thus in their survival and quality of life, can be achieved using biomarkers that facilitate improved tumor typing. We introduce and implement a combinatorial metabolic and molecular approach that applies state-of-the-art, high-resolution magic angle spinning (HRMAS) proton (1H) MRS and gene transcriptome profiling to intact brain tumor biopsies, to identify unique biomarker profiles of brain tumors. Our results show that samples as small as 2 mg can be successfully processed, the HRMAS 1H MRS procedure does not result in mRNA degradation, and minute mRNA amounts yield high-quality genomic data. The MRS and genomic analyses demonstrate that CNS tumors have altered levels of specific 1H MRS metabolites that directly correspond to altered expression of Kennedy pathway genes; and exhibit rapid phospholipid turnover, which coincides with upregulation of cell proliferation genes. The data also suggest Sonic Hedgehog pathway (SHH) dysregulation may play a role in anaplastic ganglioglioma pathogenesis. That a strong correlation is seen between the HRMAS 1H MRS and genomic data cross-validates and further demonstrates the biological relevance of the MRS results. Our combined metabolic/molecular MRS/genomic approach provides insights into the biology of anaplastic ganglioglioma and a new potential tumor typing methodology that could aid neurologists and neurosurgeons to improve the diagnosis, treatment, and ongoing evaluation of brain tumor patients.