Conference record. Asilomar Conference on Signals, Systems & Computers最新文献

It is well-known that many signals of interest can be well-estimated via just a small number of supports under some specific basis. Here, we consider finding sparse solution for Multiple Measurement Vectors (MMVs) in case of having both jointly sparse and clumpy structure. Most of the previous work for finding such sparse representations are based on greedy and sub-optimal algorithms such as Basis Pursuit (BP), Matching Pursuit (MP), and Orthogonal Matching Pursuit (OMP). In this paper, we first propose a hierarchical Bayesian model to deal with MMVs that have jointly-sparse structure in their solutions. Then, the model is modified to account for clumps of the neighbor supports (block sparsity) in the solution structure, as well. Several examples are considered to illustrate the merit of the proposed hierarchical Bayesian model compared to OMP and a modified version of the OMP algorithm.

This paper presents a method for directly estimating slope values in a noisy piecewise linear function. By imposing a Markov structure on the sequence of slopes, piecewise linear fitting is posed as a maximum a posteriori estimation problem. A dynamic program efficiently solves this by traversing a linearly growing trellis. The alternating maximization algorithm (a kind of pseudo-EM method) is used to estimate the model parameters from data and its convergence behavior is analyzed. Ultrasound shear wave imaging is presented as a primary application. The algorithm is general enough for applicability in other fields, as suggested by an application to the estimation of shifts in financial interest rate data.

Image registration is an important processing step in fluorescence microscopy, for example in tracking or super-resolution methods. Precision localization of single fluorescent molecules from a quantum limited photon detection process, subject to Gaussian readout noise, is key to the use of single molecule microscopy. It is therefore important to know the effect that registration has on the accuracy of localizing a single molecule. Here we demonstrate a suitable approach to image registration that accounts for point-wise errors in localizing the control points typically used in fluorescence microscopy. This allows expressions for the localization errors caused by the registration process to be derived, showing dependence on the number of control points and their associated photon counts.

We present recent results on the design of the RSVP Keyboard - a brain computer interface (BCI) for expressive language generation for functionally locked-in individuals using rapid serial visual presentation of letters or other symbols such as icons. The proposed BCI design tightly incorporates probabilistic contextual information obtained from a language model into the single or multi-trial event related potential (ERP) decision mechanism. This tight fusion of contextual information with instantaneous and independent brain activity is demonstrated to potentially improve accuracy in a dramatic manner. Specifically, a simple regularized discriminant single-trial ERP classifier's performance can be increased from a naive baseline of 75% to 98% in a 28-symbol alphabet operating at 5% false ERP detection rate. We also demonstrate results which show that trained healthy subjects can achieve real-time typing accuracies over 90% mostly relying on single-trial ERP evidence when supplemented with a rudimentary n-gram language model. Further discussion and preliminary results include our initial efforts involving a locked-in individual and our efforts to train him to improve his skill in performing the task.

Analysis of human brain development is a crucial step for improved understanding of neurodevelopmental disorders. We focus on normal brain development as is observed in the multimodal longitudinal MRI/DTI data of neonates to two years of age. We present a spatio-temporal analysis framework using Gompertz function as a population growth model with three different spatial localization strategies: voxel-based, data driven clustering and atlas driven regional analysis. Growth models from multimodal imaging channels collected at each voxel form feature vectors which are clustered using the Dirichlet Process Mixture Models (DPMM). Clustering thus combines growth information from different modalities to subdivide the image into voxel groups with similar properties. The processing generates spatial maps that highlight the dynamic progression of white matter development. These maps show progression of white matter maturation where primarily, central regions mature earlier compared to the periphery, but where more subtle regional differences in growth can be observed. Atlas based analysis allows a quantitative analysis of a specific anatomical region, whereas data driven clustering identifies regions of similar growth patterns. The combination of these two allows us to investigate growth patterns within an anatomical region. Specifically, analysis of anterior and posterior limb of internal capsule show that there are different growth trajectories within these anatomies, and that it may be useful to divide certain anatomies into subregions with distinctive growth patterns.

Owing to its high quantum efficiency, the charge-coupled device (CCD) is an important imaging tool employed in biological applications such as single molecule microscopy. Under extremely low light conditions, however, a CCD is generally unsuitable because its readout noise can easily overwhelm the weak signal. Instead, an electron-multiplying charge-coupled device (EMCCD), which stochastically amplifies the acquired signal to drown out the readout noise, can be used. We have previously proposed a framework for calculating the Fisher information, and hence the Cramer-Rao lower bound, for estimating parameters (e.g., single molecule location) from the images produced by an optical microscope. Here, we develop the theory that is needed for deriving, within this framework, performance measures pertaining to the estimation of parameters from an EMCCD image. Our results allow the comparison of a CCD and an EMCCD in terms of the best accuracy with which parameters can be estimated from their acquired images.

Single molecule tracking in three dimensions (3D) in a live cell environment holds the promise of revealing important new biological insights. However, conventional microscopy based imaging techniques are not well suited for fast 3D tracking of single molecules in cells. Previously we developed an imaging modality multifocal plane microscopy (MUM) to image fast intracellular dynamics in 3D in live cells. Recently, we have reported an algorithm, the MUM localization algorithm (MUMLA), for the 3D localization of point sources that are imaged using MUM. Here, we present a review of our results on MUM and MUMLA. We have validated MUMLA through simulated and experimental data and have shown that the 3D-position of quantum dots (QDs) can be determined with high spatial accuracy over a wide spatial range. We have calculated the Cramer-Rao lower bound for the problem of determining the 3D location of point sources from MUM and from conventional microscopes. Our analyses shows that MUM overcomes the poor depth discrimination of the conventional microscope, and thereby paves the way for high accuracy tracking of nanoparticles in a live cell environment. We have also shown that the performance of MUMLA comes consistently close to the Cramer-Rao lower bound.

The literature of the last three decades is replete with automatic methods for retinal image analysis. Acceptance has been limited due to post-processing or tuning requirements that may be just as time consuming as the original manual methods. The point of view herein is that by taking advantage of the human visual system and expert knowledge from the outset, the promised efficiencies of digital methods can be achieved in practice as well as in theory. Thus, simple labeling of regions of interest that is accepted and easily performed in a few moments by the human can provide enormous advantage to an already well-developed algorithm. Three examples are provided: drusen segmentation, image registration, and geographic atrophy segmentation, with applications to disease understanding.