Mukund Desai

Robust Gaussian and non-Gaussian matched subspace detection

We address the problem of matched filter and subspace detection in the presence of arbitrary noise and interference or interfering signals that may lie in an arbitrary unknown subspace of the measurement space. A minmax methodology developed to deal with this uncertainty can also be adapted to situations where partial information on the interference or other uncertainties is available. This methodology leads to a hypothesis test with adequate levels of false alarm robustness and signal detection sensitivity. The robust test is applicable to a large class of noise density functions. In addition, generalized likelihood ratio (GLR) detectors are derived for the class of generalized Gaussian noise. The detectors are generalizations of the /spl chi//sup 2/, t, and F statistics used with Gaussian noise, which are themselves motivated in a new way by the robust test. For matched filter detection, these expressions are simpler and computationally efficient. The robust test reduces to the conventional test when unlearned subspace interference is known to be absent. The results demonstrate that when compared with the conventional detector, the robust one trades off some detection performance in the absence of interference for the sake of robustness in its presence.

A fault detection and isolation methodology

A fault detection and isolation methodology has been developed for validation of sensors and plant components. Significantly, the isolation of most consistent and inconsistent subsets of measurements for the purposes of estimation and failure detection is performed on the basis of a multi-level, as opposed to the usual bi-level, fail/no fail, characterization of the inconsistencies among measurements. This is achieved by the concurrent checking of the relative consistency of smaller size subsets of measurements. The algorithm has been computer-coded for real time applications, and validated by on-line demonstration in an operating nuclear reactor.

Dual-Sensor Failure Identification Using Analytic Redundancy

In this paper we present a reliable technique for failure detection and identification for dual flight control sensors aboard the F-8 digital fly-by-wire aircraft, and we discuss the successful application of the technique to identifying failures injected on test flight downlink data. The technique exploits the analytic redundancy which exists as relationships among variables being measured by dissimilar instruments, and it accommodates both modeling errors and the allowable errors on unfailed instruments. With straightforward modification the technique may be extended to provide failure monitoring of a single remaining sensor after the identified failure of its companion sensor. Nomenclature = SPRT failure threshold ( 0) = DG case orientation angle

Paracrine and epigenetic control of trophectoderm differentiation from human embryonic stem cells: the role of bone morphogenic protein 4 and histone deacetylases.

Our understanding of paracrine and epigenetic control of trophectoderm (TE) differentiation is limited by available models of preimplantation human development. Simple, defined media for selective TE differentiation of human embryonic stem cells (hESCs) were developed, enabling mechanistic studies of early placental development. Paracrine requirements of preimplantation human development were evaluated with hESCs by measuring lineage-specific transcription factor expression levels in single cells and morphological transformation in response to selected paracrine and epigenetic modulators. Bone morphogenic protein 4 (BMP4) addition to feeder-free pluripotent stem cells on matrigel frequently formed CDX2-positive TE. However, BMP4 or activin A inhibition alone also produced a mix of mesoderm and extraembryonic endoderm under these conditions. Further, BMP4 failed to form TE from adherent hESC maintained in standard feeder-dependent monolayers. Given that the efficiency and selectivity of BMP4-induced TE depended on medium components, we developed a basal medium containing insulin and heparin. In this medium, BMP4 induction of TE was dose dependent and with activin A inhibition by SB431542 (SB), approached 100% of cells. This paracrine stimulation of pluripotent cells transformed colony morphology from a cuboidal to squamous epithelium quantitatively on day 3, and produced significant multinucleated syncytiotrophoblasts by day 8. Addition of trichostatin A, a histone deacetylase (HDAC) inhibitor, reduced HDAC3, histone H3K9 methylation, and slowed differentiation in a dose-dependent manner. Modulators of BMP4- or HDAC-dependent signaling might adversely influence the timing and viability of early blastocyst developed in vitro. Since blastocyst development is synchronized to uterine receptivity, epigenetic regulators of TE differentiation might adversely affect implantation in vivo.

Functional MRI activity characterization using response time shift estimates from curve evolution

Characterizing the response of the brain to a stimulus based on functional magnetic resonance imaging data is a major challenge due to the fact that the response time delay of the brain may be different from one stimulus phase to the next and from pixel to pixel. To enhance detectability, this work introduces the use of a curve evolution approach that provides separate estimates of the response time shifts at each phase of the stimulus on a pixel-by-pixel basis. The approach relies on a parsimonious but simple model that is nonlinear in the time shifts of the response relative to the stimulus and linear in the gains. To effectively use the response time shift estimates in a subspace detection framework, we implement a robust hypothesis test based on a Laplacian noise model. The algorithm provides a pixel-by-pixel functional characterization of the brains response. The results based on experimental data show that response time shift estimates, when properly implemented, enhance detectability without sacrificing robustness.

Analytic Redundancy for On-Line Fault Diagnosis in a Nuclear Reactor

A computer-aided diagnostic technique has been applied to on-line signal validation in an operating nuclear reactor. To avoid installation of additional redundant sensors for the sole purpose of fault isolation, a real-time model of nuclear instrumentation and the thermal-hydraulic process in the primary coolant loop was developed and experimentally validated. The model provides analytically redundant information sufficient for isolation of failed sensors as well as for detection of abnormal plant operation and component malfunctioning. Nomenclature B =bias for sensor calibration b = error bound for measurement C = specific heat F = mass flow rate of primary coolant H = measurement matrix K = product of heat transfer coefficient and area £ = number of measurements M = thermal mass m = measurement p = parity vector Q = power or rate of energy flow S = scale factor for measurement T = temperature t = time V = projection matrix v = sensor output in volts w = weighting coefficient (0 < w < 1) x = true value of a measured variable e = measurement noise 77 = parameter associated with heat transfer £ = shim blade position r = time constant X = fraction of neutron power

Performance evaluation of multiresolution texture analysis of stem cell chromatin

We apply texture image analysis to automated classification of stem cell nuclei, based on the observation that chromatin in human embryonic stem cells becomes more granular during differentiation. Using known probability models for texture multiresolution decompositions, we derive likelihood ratio test statistics. We also derive the probability density functions of these non-Gaussian statistics and use them to evaluate the performance of the classification test. Results indicate that the test can distinguish with probability 0.95 between nuclei that are pluripotent and those with varying degrees of differentiation. The test recognizes nuclei with similar differentiation level even if prior information says the contrary. This approach should be useful for classifying genome-wide epigenetic changes and chromatin remodeling during human development. Finally, the test statistics and their density functions are applicable to a general texture classification problem.

Segmentation of MR Images using Curve Evolution and Prior Information

Segmentation of anatomic structures of the human brain from MR images is important for assessing treatment efficacy, screening for anomalies, and improving our understanding of human development. The labor intensive nature of manual segmentation, however, makes such a technique viable only in selected cases. In this paper we present a new approach to segmentation that involves only minimal human interactions. The technique utilizes a variational formulation to obtain an edge-strength function over the region of interest, and uses curve evolution and a pre-segmented atlas to guide the actual segmentation process. The approach is demonstrated via both phantoms and actual MR images, and when applied to the lateral ventricles and caudate nucleus, showed a size accuracy error of 5%–20% with respect to manual segmentation, depending on the manual segmentation method utilized.

NON-INVASIVE IMAGE BASED SUPPORT VECTOR MACHINE CLASSIFICATION OF HUMAN EMBRYONIC STEM CELLS

We present a non-invasive, non-destructive automatable image-based methodology for classifying human embryonic stem cell (hESC) colonies. In contrast to differentiated colonies, pluripotent colonies contain homogeneous tight textures, thus allowing a statistical analysis of the coefficients obtained from a wavelet based texture decomposition to discriminate between the colonies. Similarly, borders of undifferentiated cell colonies are sharp, and circular, while those of differentiated colonies are not. We confine our description in this paper to texture analysis, which relies on a parametric and non-parametric hierarchical statistical classification. Parametric classification relies on probability models for texture wavelet coefficients, while non-parametric classification makes use of support vector machines. Preliminary implementation using a truth set yielded a 96% rate of successful colony classification between distant classes, while for intermediate classes of colonies, with mixed population, the success rate was at least 86%. The texture analysis was also validated using individual egg cell images

Robust Subspace Learning and Detection in Laplacian Noise and Interference

We address the problem of maximum likelihood subspace learning and detection in the presence of Laplacian noise and interference whose subspace may be known or unknown. For subspace learning, the Laplacian problem reduces to a maxmin convex mathematical program with polyhedral cost. The minimization involves projection of the measurements onto a subspace orthogonal to the signal and interference spaces and has linear constraints, while the maximization produces the subspace and has polyhedral constraints. The Laplacian noise model for subspace detection and estimation, motivated by applications in functional magnetic resonance imaging and applicable in other areas, yields maximum likelihood detectors and learned subspaces with unique structure due to the presence of corners in the polyhedrally convex optimization. For instance, the optimal learned subspace can consist of vectors whose elements take values of +1 or -1 only. Emergence of such a quantization attests to the robustness property of Laplacian learning, meaning that the solution is insensitive to perturbation in the data set. The resulting detectors are similarly robust to false alarms and have computationally attractive properties.