Squared Extrapolation Methods (SQUAREM): A New Class of Simple and Efficient Numerical Schemes for Accelerating the Convergence of the EM Algorithm

We derive a new class of iterative schemes for accelerating the convergence of the EM algorithm, by exploiting the connection between fixed point iterations and extrapolation methods. First, we present a general formulation of one-step iterative schemes, which are obtained by cycling with the extrapolation methods. We, then square the one-step schemes to obtain the new class of methods, which we call SQUAREM. Squaring a one-step iterative scheme is simply applying it twice within each cycle of the extrapolation method. Here we focus on the first order or rank-one extrapolation methods for two reasons, (1) simplicity, and (2) computational efficiency. In particular, we study two first order extrapolation methods, the reduced rank extrapolation (RRE1) and minimal polynomial extrapolation (MPE1). The convergence of the new schemes, both one-step and squared, is non-monotonic with respect to the residual norm. The first order one-step and SQUAREM schemes are linearly convergent, like the EM algorithm but they have a faster rate of convergence. We demonstrate, through five different examples, the effectiveness of the first order SQUAREM schemes, SqRRE1 and SqMPE1, in accelerating the EM algorithm. The SQUAREM schemes are also shown to be vastly superior to their one-step counterparts, RRE1 and MPE1, in terms of computational efficiency. The proposed extrapolation schemes can fail due to the numerical problems of stagnation and near breakdown. We have developed a new hybrid iterative scheme that combines the RRE1 and MPE1 schemes in such a manner that it overcomes both stagnation and near breakdown. The squared first order hybrid scheme, SqHyb1, emerges as the iterative scheme of choice based on our numerical experiments. It combines the fast convergence of the SqMPE1, while avoiding near breakdowns, with the stability of SqRRE1, while avoiding stagnations. The SQUAREM methods can be incorporated very easily into an existing EM algorithm. They only require the basic EM step for their implementation and do not require any other auxiliary quantities such as the complete data log likelihood, and its gradient or hessian. They are an attractive option in problems with a very large number of parameters, and in problems where the statistical model is complex, the EM algorithm is slow and each EM step is computationally demanding.

A Faster Circular Binary Segmentation Algorithm for the Analysis of Array CGH Data

Motivation: Array CGH technologies enable the simultaneous measurement of DNA copy number for thousands of sites on a genome. We developed the circular binary segmentation (CBS) algorithm to divide the genome into regions of equal copy number (Olshen {\it et~al}, 2004). The algorithm tests for change-points using a maximal $t$-statistic with a permutation reference distribution to obtain the corresponding $p$-value. The number of computations required for the maximal test statistic is $O(N^2),$ where $N$ is the number of markers. This makes the full permutation approach computationally prohibitive for the newer arrays that contain tens of thousands markers and highlights the need for a faster. algorithm. Results: We present a hybrid approach to obtain the $p$-value of the test statistic in linear time. We also introduce a rule for stopping early when there is strong evidence for the presence of a change. We show through simulations that the hybrid approach provides a substantial gain in speed with only a negligible loss in accuracy and that the stopping rule further increases speed. We also present the analysis of array CGH data from a breast cancer cell line to show the impact of the new approaches on the analysis of real data. Availability: An R (R Development Core Team, 2006) version of the CBS algorithm has been implemented in the ``DNAcopy'' package of the Bioconductor project (Gentleman {\it et~al}, 2004). The proposed hybrid method for the $p$-value is available in version 1.2.1 or higher and the stopping rule for declaring a change early is available in version 1.5.1 or higher.

[Acute lower gastrointestinal bleeding--an evidence-based algorithm for diagnosis and treatment]

Novel means to locate and treat lower gastrointestinal bleeding (lGB) allow to reduce the rate of required surgical interventions and help to limit the extend of resection. The risk stratification of patients with lGB is the primary step of our recommended treatment algorithm. Accordingly, risk stratifying instruments, which are only partly validated up to now, are gaining significance in lGB. Whereas, gastro-duodenoscopy and colonoscopy prior to angiography or scintigraphy are established diagnostic tools, capsule enteroscopy offers a novel approach to hemodynamic stable patients with lGB that are difficult to localize. With its every increasing sensitivity, Angio-Computer Tomography is likely to replace scintigraphy and diagnostic angiography in the very near future. In addition, recent advances in superselective microembolisation have been shown to have the potential rendering surgical interventions in a majority of patients with acute lGB unnecessary. The extend of required surgical resection is largely dependent on the success to localize the bleeding source of prior diagnostics. Only if the source is identified, a limited segmental resection should be performed. Should surgery be required, we suggest to maintain the effort to localize the bleeding, either by prior laparoscopy and/or by intraoperative entero-colonoscopy. Eventually, if the source of bleeding remains unclear total colectomy with ileorectal anastomosis represents the procedure of choice in patients with acute lGB.

Effects of expectation and caffeine on arousal, well-being, and reaction time

The objective of this study is to determine the impact of expectation associated with placebo and caffeine ingestion. We used a three-armed, randomized, double-blind design. Two three-armed experiments varying instruction (true, false, control) investigated the role of expectations of changes in arousal (blood pressure, heart rate), subjective well-being, and reaction time (RT). In Experiment 1 (N = 45), decaffeinated coffee was administered, and expectations were produced in one group by making them believe they had ingested caffeinated coffee. In Experiment 2 (N = 45), caffeinated orange juice was given in both experimental groups, but only one was informed about the true content. In Experiment 1, a significant effect for subjective alertness was found in the placebo treatment compared to the control group. However, for RT and well-being no significant effects were found. In Experiment 2, no significant expectancy effects were found. Caffeine produced large effects for blood pressure in both treatments compared to the control group, but the effects were larger for the false information group. For subjective well-being (alertness, calmness), considerable but nonsignificant changes were found for correctly informed participants, indicating possible additivity of pharmacologic effect and expectations. The results tentatively indicate that placebo and expectancy effects primarily show through introspection.

Multilevel Latent Class Models with Dirichlet Mixing Distribution

Latent class analysis (LCA) and latent class regression (LCR) are widely used for modeling multivariate categorical outcomes in social sciences and biomedical studies. Standard analyses assume data of different respondents to be mutually independent, excluding application of the methods to familial and other designs in which participants are clustered. In this paper, we develop multilevel latent class model, in which subpopulation mixing probabilities are treated as random effects that vary among clusters according to a common Dirichlet distribution. We apply the Expectation-Maximization (EM) algorithm for model fitting by maximum likelihood (ML). This approach works well, but is computationally intensive when either the number of classes or the cluster size is large. We propose a maximum pairwise likelihood (MPL) approach via a modified EM algorithm for this case. We also show that a simple latent class analysis, combined with robust standard errors, provides another consistent, robust, but less efficient inferential procedure. Simulation studies suggest that the three methods work well in finite samples, and that the MPL estimates often enjoy comparable precision as the ML estimates. We apply our methods to the analysis of comorbid symptoms in the Obsessive Compulsive Disorder study. Our models' random effects structure has more straightforward interpretation than those of competing methods, thus should usefully augment tools available for latent class analysis of multilevel data.

Collapsed cone convolution and analytical anisotropic algorithm dose calculations compared to VMC++ Monte Carlo simulations in clinical cases

The purpose of this work was to study and quantify the differences in dose distributions computed with some of the newest dose calculation algorithms available in commercial planning systems. The study was done for clinical cases originally calculated with pencil beam convolution (PBC) where large density inhomogeneities were present. Three other dose algorithms were used: a pencil beam like algorithm, the anisotropic analytic algorithm (AAA), a convolution superposition algorithm, collapsed cone convolution (CCC), and a Monte Carlo program, voxel Monte Carlo (VMC++). The dose calculation algorithms were compared under static field irradiations at 6 MV and 15 MV using multileaf collimators and hard wedges where necessary. Five clinical cases were studied: three lung and two breast cases. We found that, in terms of accuracy, the CCC algorithm performed better overall than AAA compared to VMC++, but AAA remains an attractive option for routine use in the clinic due to its short computation times. Dose differences between the different algorithms and VMC++ for the median value of the planning target volume (PTV) were typically 0.4% (range: 0.0 to 1.4%) in the lung and -1.3% (range: -2.1 to -0.6%) in the breast for the few cases we analysed. As expected, PTV coverage and dose homogeneity turned out to be more critical in the lung than in the breast cases with respect to the accuracy of the dose calculation. This was observed in the dose volume histograms obtained from the Monte Carlo simulations.

Application and assessment of a robust elastic motion correction algorithm to dynamic MRI

The purpose of this study was to assess the performance of a new motion correction algorithm. Twenty-five dynamic MR mammography (MRM) data sets and 25 contrast-enhanced three-dimensional peripheral MR angiographic (MRA) data sets which were affected by patient motion of varying severeness were selected retrospectively from routine examinations. Anonymized data were registered by a new experimental elastic motion correction algorithm. The algorithm works by computing a similarity measure for the two volumes that takes into account expected signal changes due to the presence of a contrast agent while penalizing other signal changes caused by patient motion. A conjugate gradient method is used to find the best possible set of motion parameters that maximizes the similarity measures across the entire volume. Images before and after correction were visually evaluated and scored by experienced radiologists with respect to reduction of motion, improvement of image quality, disappearance of existing lesions or creation of artifactual lesions. It was found that the correction improves image quality (76% for MRM and 96% for MRA) and diagnosability (60% for MRM and 96% for MRA).