Statistical Pattern Recognition Methods Research Articles

Examination of sol-gel production repeatability by statistical pattern recognition methods

We examine the repeatability of the sol-gel production pro- cess. Methods based on statistical pattern recognition are used to evalu- ate microscopic images of sol-gel matrices. The applied classification is a two-steps scheme: dimensionality reduction of the data and cluster analysis. Two methods of dimensionality reduction and five methods of cluster analysis are used. We show that the repeatability varies between 82 and 93.77%. The obtained results demonstrate that the molar ratio R of the number of solvent moles to the number of substrate moles influ- ences the stability of sol-gel materials, thus enabling us to reach a higher repeatability in the production process in the case of sol-gel produced with the addition of detergent. © 2003 Society of Photo-Optical Instrumentation Engineers. (DOI: 10.1117/1.1555996)

Automated segmentation of multispectral brain MR images

This work presents a robust and comprehensive approach for the in vivo automated segmentation and quantitative tissue volume measurement of normal brain composition from multispectral magnetic resonance imaging (MRI) data. Statistical pattern recognition methods based on a finite mixture model are used to partition the intracranial volume into gray matter (GM), white matter (WM), and cerebrospinal fluid (CSF) spaces. A masking algorithm initially extracts the brain volume from surrounding extrameningeal tissue. Radio frequency (RF) field inhomogeneity effects in the images are then removed using a recursive method that adapts to the intrinsic local tissue contrast. Our technique supports heterogeneous data with multispectral MR images of different contrast and intensity weighting acquired at varying spatial resolution and orientation. The proposed image segmentation methods have been tested using multispectral T1-, proton density-, and T2-weighted MRI data from young and aged non-human primates as well as from human subjects.

Analyzing the physiological signature of anabolic steroids in cattle urine using pyrolysis/metastable atom bombardment mass spectrometry and pattern recognition.

Pyrolysis coupled to metastable atom bombardment (MAB) and time-of-flight mass spectrometry (TOFMS) is used for generating mass spectra from bovine urine samples obtained from cattle treated with anabolic steroids. These spectra constitute fingerprints, which can be discriminated by multivariate statistical analysis. Four main conclusions can be drawn from this work: (i) The use of different metastable gases, such as Xe*, Kr*, or N2*, as an energy-tunable ionization beamline allows control of the internal energy and the dissociation processes of the produced odd electron molecular ions, thus giving rise to complementary mass spectra fingerprints. (ii) A variable transformation depending on the biofluid matrix suitably contracts the frequency distribution of the generated data for low m/z ratios holding information related to endogenous metabolites encountered in urine. (iii) Coupling variable selection to statistical pattern recognition methods results in low error rates (< 1%) for predicting MAB mass fingerprints, especially using lineardiscriminant analysis (LDA). (iv) LDA discriminates controls from treated animals and also correlates to quantitative physiological responses induced by anabolic steroids. This work shows that Py-MAB-TOFMS could be a suitable method for complementary monitoring anabolic use in sports, medicine, and cattle breeding, as well as monitoring many other long-lasting although weak physiological disruptions.

Quantitative analysis of NO2 in the presence of CO using a single tungsten oxide semiconductor sensor and dynamic signal processing.

We demonstrate that NO2 can be quantitatively analysed in the presence of CO using a single tungsten oxide based resistive gas sensor. The working temperature of the sensor was modulated between 190 and 380 degrees C and its dynamic response to different concentrations of CO, NO2, and CO + NO2 mixtures was monitored. Either the fast Fourier transform (FFT) or the discrete wavelet transform (DWT) was used to extract important features from the sensor response. These features were then input to different (statistical and neural) pattern recognition methods. The species considered can be discriminated with a success rate higher than 90% using a Fuzzy ARTMAP or a radial basis function neural network. The concentrations of the gases studied can be accurately predicted, by using the DWT coupled to partial least squares (PLS) models. The correlation coefficients of the predicted versus real concentrations were 0.923, 0.870 and 0.866 for CO, NO2, and NO2 in CO + NO2 mixtures, respectively.

Chemical profiling to differentiate geographic growing origins of coffee.

The objective of this research was to demonstrate the feasibility of this method to differentiate the geographical growing regions of coffee beans. Elemental analysis (K, Mg, Ca, Na, Al, V, Cr, Mn, Fe, Co, Ni, Cu, Zn, Mo, S, Cd, Pb, and P) of coffee bean samples was performed using ICPAES. There were 160 coffee samples analyzed from the three major coffee-growing regions: Indonesia, East Africa, and Central/South America. A computational evaluation of the data sets was carried out using statistical pattern recognition methods including principal component analysis, discriminant function analysis, and neural network modeling. This paper reports the development of a method combining elemental analysis and classification techniques that may be widely applied to the determination of the geographical origin of foods.

Identifying the 3'-terminal exon in human DNA.

We present JTEF, a new program for finding 3' terminal exons in human DNA sequences. This program is based on quadratic discriminant analysis, a standard non-linear statistical pattern recognition method. The quadratic discriminant functions used for building the algorithm were trained on a set of 3' terminal exons of type 3tuexon (those containing the true STOP codon). We showed that the average predictive accuracy of JTEF is higher than the presently available best programs (GenScan and Genemark.hmm) based on a test set of 65 human DNA sequences with 121 genes. In particular JTEF performs well on larger genomic contigs containing multiple genes and significant amounts of intergenic DNA. It will become a valuable tool for genome annotation and gene functional studies. JTEF is available free for academic users on request from ftp://cshl.org/pub/science/mzhanglab/JTEF and will be made available through the World Wide Web (http://argon.cshl.org/).

Following the recovery of naso-pharyngeal cancer patients by trace elements in hair using statistical pattern recognition methods.

This article describes a study where the recovery of nasopharyngeal cancer (NPC) patients was traced by trace elements (TEs) in hair of these patients. These patients whose concentrations of TEs in hair were studied are volunteers who were divided into three groups B, C, and D. Group B was made up of volunteers who had just been diagnosed as having NPC, group C and group D were made up of volunteers who had been diagnosed as having NPC after 3 and 6 mo, respectively, of treatment. For comparison, a control group, namely group A, which was made up of volunteers from healthy persons was added to this study. By implementation of statistic pattern recognition methods, it has been found that the concentrations of TEs in hair can remarkably reflect different recovery phases of NPC patients.

New optical measuring system for solder joint inspection

Due to specularity and complex 3D geometry, the visual inspection of solder joints has been regarded as one of the most difficult tasks and thus has not guaranteed accurate inspection results. This paper deals with an optical sensing system designed to inspect solder joints automatically. Taking into account a specular characteristic, a new optical solder joint inspection system is proposed to obtain the external profile of solder joints accurately. A laser scanning unit scans the area of solder joint and observes the angle of the reflected beam. To realize this measuring principle, the system is so composed that a galvanometer steers the direction of laser beam to a point to be measured, a mirror unit (a parabolic and a conic mirror) gathers the specular components of the reflected beam toward its center and a beam-receiving unit positioned along the mirror unit's center line detects the reflected beam. To classify the defects of solder joint, a statistical pattern recognition method is utilized. To verify the validity of the developed system, a series of experiments was performed for SOPs and QFPs in insufficient, normal and excess soldering condition. Based upon observation of the experimental results, the proposed system is found to show good performance for inspection of solder joint defects.

Modeling the student in intelligent tutoring sytems: The promise of a new psychometrics

This paper reviews a number of relatively new and promising psychometric approaches to the problem of modeling student achievement (the student model) within intelligent tutoring systems (ITS). A shared characteristic of most ITSs is their need to estimate a model of the student's understanding of the domain, and use this model to modify and adapt subsequent instructional content and sequence. Sound cognitive diagnosis and the need to advance ITS technology require the development of student models that are integrated with cognitive theory and instructional science. A number of cognitively oriented psychometric approaches — including latent-trait models, statistical pattern recognition methods, and causal probabilistic networks — are described and discussed within the current ITS framework. As measurement-based student models are refined, we anticipate their compatibility with future generations of intelligent tutoring systems.

Model-Free Texture Segmentation Based on Distances between First-Order Statistics

This work describes several new image segmentation approaches, all based on statistical edge detection. Although the involved statistical pattern recognition methods are well established, their adaptation to an edge-oriented segmentation is an extremely uncommon topic in the image processing literature. Thus, the subject of this paper is both tutorial and innovative in nature, with respect to pattern recognition and to image segmentation, respectively. In this approach, the borders between homogeneous regions are detected by evaluating a "measure of diversity" between symmetrical and equal-sized subsets of the observation window. Each neighborhood is characterized by its complete local gray value distribution. In case the characterization by the first-order statistics is not sufficient, information on the gray values′ spatial relationships must be extracted from the second order statistics. The work focuses on the analysis of efficient methods for measuring a "degree of diversity" following several approaches, namely: • Distances between the vectors of the rank-ordered gray values: Minkowski-, Canberra-, Tanimoto-distance, scalar product, composite distance (newly developed here). • Distances between estimated gray value density functions: distances of Kolmogorov, Bhattacharyya, and Patrick-Fischer. The estimation is performed by means of Parzen kernels. • Measures of cluster separation between subwindows taken in the one-dimensional gray value space or in the two-dimensional space obtained by also considering a spatial distribution feature. • Nonparametric Wilcoxon-type two-population tests. • Distances between the estimated locations and between the estimated scales of symmetrical subwindows. Experimental results obtained on natural images with problematic textures (remote sensing, radar, ultrasound, nuclear medicine) and on synthetic images, with all the approaches mentioned above, are shown and illustrated.

Temporal, Spatial, and Environmental classification of Pine Reflectance Spectra

To create consistent spectral data for evaluating geographical, chronological, and environmental differences or events in main Scots pine populations in Finland and Sweden, reflectance spectra from randomly sampled single pine individuals (N = 2069) were measured. The spectra were recorded in 1991-1993 during and after the growing season by using a mobile spectroradiometer working in the 390-1070 nm wavelength range. The spectral data was processed by using three independent statistical pattern recognition methods. Typical spectral shape differences between the growing seasons were observed due to the obvious phenological phase shift from year to year. This was verified by the observed differences inside the growing season. Longitudinal distinguishing characteristics of the pine spectra were greater than latitudinal ones. We are also able to classify pines growing in polluted and less polluted environments to the corresponding classes.

Comparative analysis of statistical pattern recognition methods in high dimensional settings

An extensive simulation study is reported comparing eight statistical classification methods, focusing on problems where the number of observations is less than the number of variables. Using a wide range of artificial and real data sets, two types of classifiers are contrasted; methods that classify using all variables, and methods that first reduce the number of dimensions to two or three. The simulations identified regularized discriminant analysis as the overall clearly most powerful classifier, and show that in most cases, a reduction of the dimensionality to two or three dimensions prior to classification increases the error in allocating test observations.

Classification of the reflectance spectra of pine, spruce, and birch

Statistical pattern-recognition methods are applied to the classification of the reflectance spectra of growing trees (Scots pine, Norway spruce, and birch). We show by using large training sets that it is possible to develop classification filters that are able to discriminate the tree types with a very high probability. Our approach may offer a reference coordinate system for multispectral remote sensing of different levels of forest damage.

Prediction-based dynamic load-sharing heuristics

Presents dynamic load-sharing heuristics that use predicted resource requirements of processes to manage workloads in a distributed system. A previously developed statistical pattern-recognition method is employed for resource prediction. While nonprediction-based heuristics depend on a rapidly changing system status, the new heuristics depend on slowly changing program resource usage patterns. Furthermore, prediction-based heuristics can be more effective since they use future requirements rather than just the current system state. Four prediction-based heuristics, two centralized and two distributed, are presented. Using trace driven simulations, they are compared against random scheduling and two effective nonprediction based heuristics. Results show that the prediction-based centralized heuristics achieve up to 30% better response times than the nonprediction centralized heuristic, and that the prediction-based distributed heuristics achieve up to 50% improvements relative to their nonpredictive counterpart.< <ETX xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">&gt;</ETX>

New signal processing techniques for the decomposition of EMG signals.

This paper relates to the use of knowledge-based signal processing techniques in the decomposition of EMG signals. The aim of the research is to automatically decompose EMG signals recorded at force levels up to 20 per cent maximum voluntary contraction (MVC) into their constituent motor unit action potentials (MUAPS), and to display the MUAP shapes and firing times for the clinician. This requires the classification of nonoverlapping MUAPs and superimposed waveforms formed from overlapping MUAPs in the signal. Nonoverlapping MUAPs are classified using a statistical pattern-recognition method. The decomposition of superimposed waveforms uses a combination of procedural and knowledge-based methods. The decomposition method was tested on real and simulated EMG data recorded at force levels up to 20 per cent MVC. The different EMG signals contained up to six motor units (MUs). The new decomposition program classifies the total number of MUAP firings in an EMG signal with an accuracy always greater than 95 per cent. The decomposition program takes about 15s to classify all nonoverlapping MUAPs in EMG signal of length 1.0s and, on average, an extra 9s to classify each superimposed waveform.

Comparison of major and trace components of Chinese and Canadian soils

INAA was used to determine 32 major and trace elements in 21 soil samples collected from the Loess Plateau, China, and Ontario, Canada. The elemental concentrations determined in different layers of each soil, have been used for comparison in order to find out the general characteristics and any significant differences between Canadian and Chinese soils which developed from similar parent materials. This data can be useful to pedologists and geochemists to study the elemental behaviours and the influences of climate and environment on soil formation. Effort is directed to applying a statistical factor pattern recognition method to these soils for possible use in future soil erosion and transport studies in China. From this study, four soil factors were resolved each for the Chinese and Canadian soils.

Color classification by vector subspace method and its optical implementation using liquid crystal spatial light modulator

Color classification by a statistical pattern recognition method, the subspace method, was implemented by a simple optical system including a liquid crystal spatial light modulator. Area coding filters corresponding to eigenvectors were drawn on the liquid crystal panel. The vector inner products between the filter functions of each class and the input spectral distributions were calculated instantaneously. To construct low-dimensional subspaces for each class, the recognition system for arbitrary spectral distributions was used and efficient classification filters were constructed after several iterative learning cycles. Samples with small hue differences were correctly classified.

Characterization of Yeast Strains of the Genera Candida, Hansenula, Kluyveromyces, Pichia, Rhodotorula, and Saccharomyces by Mixed-Dye Fluorometry

An analytical method in which we used the selective adsorption of several fluorophores by yeast cells is described. The suitability of using binary mixtures of 1-pyrene butyric acid, 3,6-dimethylamino acridine, 4-acetamido-4'-isothiocyanatostilbene-2,2'-disulfonic acid, and rhodamine B isothiocyanate for the characterization and identification of microorganisms was tested with 98 yeast strains belonging to the genera Candida, Hansenula, Kluyveromyces, Pichia, Rhodotorula, and Saccharomyces. The application of multivariate statistical methods and pattern recognition methods to the allocation of the yeast strains into genus-species-strain structures and to a comparison of fluorescence data sets for differentiation and identification purposes showed the usefulness of the method.

An Application of Unsupervised Neural Network Methodology Kohonen Topology‐Preserving Mapping to QSAR Analysis

AbstractThe concept and methodology of artificial neural networks is introduced. Like pattern recognition, the techniques can be classified as supervised (requiring a priori knowledge of class membership) and unsupervised (making no assumptions about class membership). An unsupervised neural network method, Kohonen Topology‐Preserving Mapping, is applied to a wide matrix of physicochemical property data for a set of antifilarial antimycin analogues containing structural outliers. Principal component analysis failed to give a good 2D representation of the data set as a whole due to linear constraints in the model which gave undue influence to the outliers. Kohonen mapping compared favourably with non‐linear unsupervised statistical pattern recognition methods for 2D representation of compound similarity and for classification based on antifilarial activity. It may prove a valuable technique for QSAR in situations where a linear method does not model the data well and a high throughput of test compounds is indicated.

Supervised Pattern Recognition Techniques in Quantitative Diagnostic Ultrasound

The methods of statistical pattern recognition are well suited to the problems of in vivo ultrasonic tissue characterization. This paper describes supervised pattern recognition methods for selecting features for tissue classification, calculating decision boundaries within the selected feature space, and evaluating the performance. We address the considerations of dimensionality and feature size which are important in classification problems where the underlying probability distributions are not completely known. Examples are given for the detection of diffuse liver disease in the clinical environment.