Johan Lim

Image Segmentation Using Hidden Markov Gauss Mixture Models

Image segmentation is an important tool in image processing and can serve as an efficient front end to sophisticated algorithms and thereby simplify subsequent processing. We develop a multiclass image segmentation method using hidden Markov Gauss mixture models (HMGMMs) and provide examples of segmentation of aerial images and textures. HMGMMs incorporate supervised learning, fitting the observation probability distribution given each class by a Gauss mixture estimated using vector quantization with a minimum discrimination information (MDI) distortion. We formulate the image segmentation problem using a maximum a posteriori criteria and find the hidden states that maximize the posterior density given the observation. We estimate both the hidden Markov parameter and hidden states using a stochastic expectation-maximization algorithm. Our results demonstrate that HMGMM provides better classification in terms of Bayes risk and spatial homogeneity of the classified objects than do several popular methods, including classification and regression trees, learning vector quantization, causal hidden Markov models (HMMs), and multiresolution HMMs. The computational load of HMGMM is similar to that of the causal HMM.

Concomitants of Multivariate Order Statistics With Application to Judgment Poststratification

We generalize the definition of a concomitant of an order statistic in the multivariate case, develop general expressions for its density, and establish related properties. We study the concomitant of a normal random vector in detail and discuss methods for calculating its moments. Furthermore, we apply the theory to develop new estimators of the mean from a judgment poststratified sample, where poststrata are formed by rank classes of auxiliary variables. Our estimators are shown to be more efficient than existing ones and robust against violations of the normality assumption. They are also well suited to applications requiring cost efficiency.

Method to classify dental arch forms

INTRODUCTION The aim of this study was to propose a method to classify dental arch forms of subjects with normal occlusion into several types that can ensure both goodness of fit and clinical application. METHODS We selected 306 subjects with normal occlusion from 15,836 young adults, recorded 14 reference points that defined the distance between 2 arch forms as the area between 2 arches, and then classified the dental arch forms by using the partitioning around medoids clustering and silhouette method. We measured tooth size, arch width, basal arch width, arch depth, mesiodistal angulations, and buccolingual inclinations. RESULTS We identified 3 types of arch forms, and cross-classification of the maxillary by mandibular arch forms showed a more frequent distribution in the diagonal elements than in the off-diagonal elements. The 3 arch forms showed differences in tooth size, arch width, basal arch width, and inclination of the posterior teeth. CONCLUSIONS By defining area discrepancies as distance measures and applying them to the cluster method by using medoids, the dental arch form can be classified keeping control for the extremes without bias. It is hoped that this method will have possible clinical applications in determining the shape and number of preformed orthodontic arch forms.

A Nonparametric Mean Estimator for Judgment Poststratified Data

MacEachern, Stasny, and Wolfe (2004, Biometrics60, 207-215) introduced a data collection method, called judgment poststratification (JPS), based on ideas similar to those in ranked set sampling, and proposed methods for mean estimation from JPS samples. In this article, we propose an improvement to their methods, which exploits the fact that the distributions of the judgment poststrata are often stochastically ordered, so as to form a mean estimator using isotonized sample means of the poststrata. This new estimator is strongly consistent with similar asymptotic properties to those in MacEachern et al. (2004). It is shown to be more efficient for small sample sizes, which appears to be attractive in applications requiring cost efficiency. Further, we extend our method to JPS samples with imprecise ranking or multiple rankers. The performance of the proposed estimators is examined on three data examples through simulation.

Effects of storage period and heat treatment on phenolic compound composition in dried Citrus peels (Chenpi) and discrimination of Chenpi with different storage periods through targeted metabolomic study using HPLC-DAD analysis

Dried Citrus peels, known as Chenpi in Chinese medicine, are a traditional medicine for the treatment of indigestion and inflammatory syndromes. In this study, we evaluated the effect of storage periods (1-year vs. 3-year) and heat treatment (90min vs. 3h at 120°C) on the total phenolic content (TPC) and bioactivity (anti-oxidant activity) of Chenpi. It was found that the long-term stored Chenpi had a higher TPC and superior 1,1-diphenyl-2-picrylhydrazyl (DPPH) radical-scavenging activity levels compared to the regular stored Chenpi, and that heat treatment increased both TPC and bioactivity. Subsequently, we developed and validated a high performance liquid chromatography with diode-array detection (HPLC-DAD) method to determine individual phenolic acids. Eleven phenolic compounds were determined in different Chenpi samples. Concentrations of total phenolic compounds were higher in long-term stored Chenpi and heat treatment raised the levels of those in regular stored Chenpi. In further study, a targeted metabolomic approach was applied to discriminate Chenpi with different storage periods. Two different phenolic acid fractions (free and ester) from the regular and long-term stored Chenpi were analyzed using the developed HPLC-DAD and the data were used in principal component analysis (PCA) on the HPLC-DAD peak areas of the 11 phenolic acids. Two principal components (PC1 and PC2) accounted for 87.1% of the variation between the regular and long-term stored Chenpi. In a two-dimensional plot of PC1 and PC2, the samples divided into four groups: two storage periods and two fractions.

Determination of bioactive compounds in fermented soybean products using GC/MS and further investigation of correlation of their bioactivities

The active ingredients and bioactivities (anti-oxidant, anti-tyrosinase, anti-proliferative and estrogenic activities) of soybean and soybean products (Cheonggukjang, Meju, Makjang, Doenjang and soy sauce) produced by different fermentation processes were compared. There were high correlations between active ingredients and bioactivities. Free phenolic acids extracted from soybean products were identified and quantified by gas chromatography/mass spectrometry (GC/MS). Overall, the components and activities in fermented soybean products were different than those in soybeans. Total phenolic content (TPC), protein content (PC) and anti-oxidant activity increased as fermentation time increased. TPC and PC showed strong negative correlations with anti-oxidant activity. Doenjang and soy sauce, two long-term fermented products, showed lower total flavonoid content (TFC) and estrogenic activities than short-term fermented soybean products. This might be explained by the decomposition and hydrolysis of flavonoids due to the long fermentation time and high temperature. Strong anti-proliferative activity against cancer cell lines, which was highly correlated with TFC, was found in Meju and Cheonggukjang. Soybean and all fermented products except Meju exhibited effective tyrosinase inhibitory activities. Fermented products showed stronger estrogenic activity than soybeans, which was highly correlated with syringic acid.

A Numerical Study of PQL Estimation Biases in Generalized Linear Mixed Models Under Heterogeneity of Random Effects

The penalized quasi-likelihood (PQL) approach is the most common estimation procedure for the generalized linear mixed model (GLMM). However, it has been noticed that the PQL tends to underestimate variance components as well as regression coefficients in the previous literature. In this article, we numerically show that the biases of variance component estimates by PQL are systematically related to the biases of regression coefficient estimates by PQL, and also show that the biases of variance component estimates by PQL increase as random effects become more heterogeneous.

Distinguishing between genotoxic and non-genotoxic hepatocarcinogens by gene expression profiling and bioinformatic pathway analysis

A rapid and sensitive method to determine the characteristics of carcinogens is needed. In this study, we used a microarray-based genomics approach, with a short-term in vivo model, in combination with insights from statistical and mechanistic analyses to determine the characteristics of carcinogens. Carcinogens were evaluated based on the different mechanisms involved in the responses to genotoxic carcinogens and non-genotoxic carcinogens. Gene profiling was performed at two time points after treatment with six training and four test carcinogens. We mapped the DEG (differentially expressed gene)-related pathways to analyze cellular processes, and we discovered significant mechanisms that involve critical cellular components. Classification results were further supported by Comet and Micronucleus assays. Mechanistic studies based on gene expression profiling enhanced our understanding of the characteristics of different carcinogens. Moreover, the efficiency of this study was demonstrated by the short-term nature of the animal experiments that were conducted.

Texture classification based on multiple Gauss mixture vector quantizers

We propose a texture classification method using multiple Gauss mixture vector quantizers (GMVQ). We designed a separate model codebook or Gauss mixture for each texture using the generalized Lloyd algorithm with a minimum discrimination information (MDI) distortion based on a training data set. The multi-codebook structure of the GMVQ classifier is an extension to images of the isolated utterance speech recognizer of J.E. Shore and D. Burton (see Proc. Int. Conf. Acoust., Speech, and Sig. Processing, IEEE82Ch.1746-7, p.907-10, 1982). We applied the algorithm to the Brodatz texture database and showed it to be competitive in performance in comparison to other texture classifiers. Its low complexity implementation and real-time operation make the approach suitable for content-based image retrieval.

Variation of the intermaxillary tooth-size relationship in normal occlusion

The purpose of this study was to explore the intermaxillary tooth-size relationship that is attributed to normal occlusion using multivariate cluster analysis, while simultaneously incorporating the full dentition as a data set. From the central incisor to the second molar, the tooth sizes of 307 subjects (188 males and 119 females; mean age ± standard deviation, 19.9 ± 3.3 years) with normal occlusion were investigated. Tooth-size data were analysed separately for the maxilla and the mandible. When clustering, the partitioning around medoids (PAM) algorithm was performed with the transformed data based on principal component analysis (PCA). After the subjects were classified into four groups, the cluster memberships were cross-classified, and the distribution pattern and intermaxillary tooth-size relationships were explored. Bolton tooth ratio showed a relatively wide range, and this was indicative of the variability in tooth size in subjects with a normal occlusion. However, the patterns of the intermaxillary tooth-size relationship were similar for males and females, and this result was concordant with the findings of the classic Bolton analysis. Using the multivariate approach to analyse the tooth-size data set of an individual patient and then comparing the results with the normal occlusion cluster has possible clinical applications in determining the amount and location of tooth-size control in orthodontics.