Pattern Recognition Tools Research Articles

Real-Time Angular Velocity-Based Misfire Detection Using Artificial Neural Networks

In this paper, an artificial neural network (ANN) is introduced in order to detect the occurrence of misfire in an internal combustion (IC) engine by analyzing the crankshaft angular velocity. This study presents three reliable misfire detection procedures. In the first two methods, the fault features are extracted using both time domain and frequency domain techniques, and a multilayer perceptron (MLP) serves as the pattern recognition tool for detecting the misfiring cylinder. In the third method, a one-dimensional (1D) convolutional neural network (CNN) that combines feature extraction capability and pattern recognition is adopted for misfire detection. The experimental data are obtained by setting a six in-line diesel engine with different cylinder misfiring to work under representative operating conditions. Finally, all three diagnostic methods achieved satisfactory results, and the 1D CNN achieved the best performance. The current study provides a novel way to detect misfiring in IC engines.

FlatDTW – Dynamic Time Warping optimization for piecewise constant templates

The aim of this work is to construct and analyze a method of optimization of the Dynamic Time Warping (DTW) algorithm – a well-known and popular pattern recognition tool. DTW is typically used to evaluate the degree of similarity between time series subjected to non-linear time warping, which is a common problem i.a. in speech recognition, music information retrieval, motion analysis and gesture recognition. Based on the dynamic programming principle, DTW is computed efficiently in O(N2) time, which contributes to its popularity and makes it applicable in many real-time scenarios. In this paper we propose a method of DTW algorithm optimization in the case when one of the time series may be modeled as a piecewise constant sequence. Such sequences, composed of several “flat” sections, typically occur when real-world data are compared to some predefined templates, as in the task of matching a melody sung by a user to a MIDI-based template melody (a.k.a. Query-by-Singing/Humming, QbH). The modified DTW processes such templates in a more efficient way reducing the computational complexity. The obtained speed-up of the algorithm is not associated with any quality loss of the matching process results.

Deep depth-based representations of graphs through deep learning networks

Graph-based representations are powerful tools in structural pattern recognition and machine learning. In this paper, we propose a framework of computing the deep depth-based representations for graph structures. Our work links the ideas of graph complexity measures and deep learning networks. Specifically, for a set of graphs, we commence by computing depth-based representations rooted at each vertex as vertex points. In order to identify an informative depth-based representation subset, we employ the well-known k-means method to identify M dominant centroids of the depth-based representation vectors as prototype representations. To overcome the burdensome computation of using depth-based representations for all graphs, we propose to use the prototype representations to train a deep autoencoder network, that is optimized using Stochastic Gradient Descent together with the Deep Belief Network for pretraining. By inputting the depth-based representations of vertices over all graphs to the trained deep network, we compute the deep representation for each vertex. The resulting deep depth-based representation of a graph is computed by averaging the deep representations of its complete set of vertices. We theoretically demonstrate that the deep depth-based representations of graphs not only reflect both the local and global characteristics of graphs through the depth-based representations, but also capture the main structural relationship and information content over all graphs under investigations. Experimental evaluations demonstrate the effectiveness of the proposed method.

Identifying channel sand-body from multiple seismic attributes with an improved random forest algorithm

Machine learning provides numerous data-driven tools for automatic pattern recognition. Even though various algorithms such as neural networks and support vector machines have been widely applied, it is still necessary to explore new paradigms and algorithms to improve the machine learning assisted seismic interpretation. Random Forest (RF) is a widely used ensemble algorithm, however, only limited studies of random forest in the seismic application were published. In this article, the methodology of random forest is introduced systematically. Meanwhile, to solve the problem of hyper-parameter determination, we propose an improved algorithm named Pruning Random Forest (PRF). To reveal the advantages of PRF in terms of predictive performance, robustness, and feature selection compared with support vector machine, neural network, and decision tree, several well-designed experiments are executed based on the seismic data of the western Bohai Sea of China. The potential and advantages of random forest in the present case are confirmed by various experiments, which substantiates that the proposed pruning random forest algorithm provides a reliable alternative way for further machine learning assisted seismic interpretation.

SAR Target Configuration Recognition via Product Sparse Representation.

Sparse representation (SR) has been verified to be an effective tool for pattern recognition. Considering the multiplicative speckle noise in synthetic aperture radar (SAR) images, a product sparse representation (PSR) algorithm is proposed to achieve SAR target configuration recognition. To extract the essential characteristics of SAR images, the product model is utilized to describe SAR images. The advantages of sparse representation and the product model are combined to realize a more accurate sparse representation of the SAR image. Moreover, in order to weaken the influences of the speckle noise on recognition, the speckle noise of SAR images is modeled by the Gamma distribution, and the sparse vector of the SAR image is obtained from q statistical standpoint. Experiments are conducted on the moving and stationary target acquisition and recognition (MSTAR) database. The experimental results validate the effectiveness and robustness of the proposed algorithm, which can achieve higher recognition rates than some of the state-of-the-art algorithms under different circumstances.

Feed-Forward Back-Propagation Neural Networks in Side-Channel Information Characterization

The safety of cryptosystems, mainly based on algorithmic improvement, is still vulnerable to side-channel attacks (SCA) based on machine learning. Multi-class classification based on neural networks and principal components analysis (PCA) can be powerful tools for pattern recognition and classification of side-channel information. In this paper, an experimental investigation was conducted to explore the efficiency of various architectures of feed-forward back-propagation (FFBP) neural networks and PCA against side-channel attacks. The experiment is performed on the data leakage of an FPGA implementation of elliptic curve cryptography (ECC). Our results show that the proposed method is a promising method for SCA with an overall accuracy of 88% correct classification.

Face-Detection Algorithm Handles Big Data To Help Identify Candidates for Restimulation

This article, written by Special Publications Editor Adam Wilson, contains highlights of paper SPE 187328, “From Face Detection to Fractured-Reservoir Characterization: Big Data Analytics for Restimulation-Candidate Selection,” by Egbadon Udegbe, SPE, Eugene Morgan, SPE, and Sanjay Srinivasan, SPE, The Pennsylvania State University, prepared for the 2017 SPE Annual Technical Conference and Exhibition, San Antonio, Texas, USA, 9–11 October. The paper has not been peer reviewed. The recent proliferation of subsurface data from instrumented wells has created significant challenges for traditional production-data-analysis methods to extract useful information for reservoir management. This paper demonstrates the viability of a production-data-classification approach adapted from real-time face detection for identifying restimulation candidates. The approach has the potential to be used as a big-data analytic tool for long-duration production-data analysis to serve as a screening tool for selection of restimulation candidates. Introduction Restimulation treatments in producing shale wells have the potential to improve economic performance by increasing the conductivity of existing fractures or enhancing their contact with the formation. The influence of matrix and fracture characteristics on the success of restimulation, however, is not completely understood, which has led to uncertainty in determining favorable candidate wells. Several methods to select restimulation candidates have been proposed. These methods, however, are time-consuming and tend to require detailed input data or exhibit a lack of generality for other reservoir settings. This paper aims to address these challenges with a new methodology for fast and robust analysis of production data from hydraulically fractured wells. A dual-permeability forward-flow modeling approach is used to generate multiple realizations of production-rate profiles by modifying fracture and other parameters. Using this data, pattern-recognition tools are applied to help uncover trends associated with favorable and unfavorable restimulation candidates. This is achieved using a binary classification framework adapted from real-time face detection, which uses simple numerical criteria computed directly from raw flow-rate data, thus eliminating the need for detailed information and promoting computational efficiency. The algorithm also provides probabilistic predictions, which serve as a means to rank candidate wells. While the process of training the classifier has the potential to be computationally intensive, the application of the trained classifier on the observed data is extremely fast, making the method useful for real-time classification of well performance. Face-Detection Overview The Viola-Jones face-detection algorithm is a binary classification scheme that takes labeled examples of images containing faces and nonfaces and develops a set of numerical criteria for distinguishing between both categories. Using these rules, an arbitrary test image can be analyzed and classified as a face or a nonface.

Using machine learning to classify extant apes and interpret the dental morphology of the chimpanzee-human last common ancestor

Author(s): Monson, Tesla A.; Armitage, David W.; Hlusko, Leslea J. | Abstract: Machine learning is a formidable tool for pattern recognition in large datasets. We developed and expanded on these methods, applying machine learning pattern recognition to a problem in paleoanthropology and evolution. For decades, paleontologists have used the chimpanzee as a model for the chimpanzee-human last common ancestor (LCA) because they are our closest living primate relative. Using a large sample of extant and extinct primates, we tested the hypothesis that machine learning methods can accurately classify extant apes based on dental data. We then used this classification tool to observe the affinities between extant apes and Miocene hominoids. We assessed the discrimination accuracy of supervised learning algorithms when tasked with the classification of extant apes (n=175), using three types of data from the postcanine dentition: linear, 2-dimensional, and the morphological output of two genetic patterning mechanisms that are independent of body size: molar module component (MMC) and premolar-molar module (PMM) ratios. We next used the trained algorithms to classify a sample of fossil hominoids (n=95), treated as unknowns. Machine learning classifies extant apes with greater than 92% accuracy with linear and 2-dimensional dental measurements, and greater than 60% accuracy with the MMC and PMM ratios. Miocene hominoids are morphologically most similar in dental size and shape to extant chimpanzees. However, relative dental proportions of Miocene hominoids are more similar to extant gorillas and follow a strong trajectory through evolutionary time. Machine learning is a powerful tool that can discriminate between the dentitions of extant apes with high accuracy and quantitatively compare fossil and extant morphology. Beyond detailing applications of machine learning to vertebrate paleontology, our study highlights the impact of phenotypes of interest and the importance of comparative samples in paleontological studies.

Logic Computing with Stateful Neural Networks of Resistive Switches.

Brain-inspired neural networks can process information with high efficiency, thus providing a powerful tool for pattern recognition and other artificial intelligent tasks. By adopting binary inputs/outputs, neural networks can be used to perform Boolean logic operations, thus potentially surpassing complementary metal-oxide-semiconductor logic in terms of area efficiency, execution time, and computing parallelism. Here, the concept of stateful neural networks consisting of resistive switches, which can perform all logic functions with the same network topology, is introduced. The neural network relies on physical computing according to Ohm's law, Kirchhoff 's law, and the ionic migration within an output switch serving as the highly nonlinear activation function. The input and output are nonvolatile resistance states of the devices, thus enabling stateful and cascadable logic operations. Applied voltages provide the synaptic weights, which enable the convenient reconfiguration of the same circuit to serve various logic functions. The neural network can solve all two-input logic operations with just one step, except for the exclusive-OR (XOR) needing two sequential steps. 1-bit full adder operation is shown to take place with just two steps and five resistive switches, thus highlighting the high efficiencies of space, time, and energy of logic computing with the stateful neural network.

A probabilistic model for state sequence analysis in hidden Markov model for hand gesture recognition

AbstractThe role of gesture recognition is significant in areas like human‐computer interaction, sign language, virtual reality, machine vision, etc. Among various gestures of the human body, hand gestures play a major role to communicate nonverbally with the computer. As the hand gesture is a continuous pattern with respect to time, the hidden Markov model (HMM) is found to be the most suitable pattern recognition tool, which can be modeled using the hand gesture parameters. The HMM considers the speeded up robust feature features of hand gesture and uses them to train and test the system. Conventionally, the Viterbi algorithm has been used for training process in HMM by discovering the shortest decoded path in the state diagram. The recursiveness of the Viterbi algorithm leads to computational complexity during the execution process. In order to reduce the complexity, the state sequence analysis approach is proposed for training the hand gesture model, which provides a better recognition rate and accuracy than that of the Viterbi algorithm. The performance of the proposed approach is explored in the context of pattern recognition with the Cambridge hand gesture data set.

Tracing university–industry knowledge transfer through a text mining approach

This study investigates knowledge transfer of university research to industry moving forward from traditional indicators by using methods from computational linguistics. We introduce a novel empirical use of pattern recognition and text mining tools to compare scientific publications to company documents. The contribution of the paper is twofold; first, a new method for tracing knowledge transfer is suggested and, second, our understanding of university–industry knowledge transfer is increased by introducing an additional perspective. We find that common text mining tools are suitable to identify concrete chunks of research knowledge within the collaborating industry. The method proves direct links between published university research and the information disclosed by companies in their websites and documents. We offer an extension to commonly used concepts, which rely either on qualitative case studies or the assessment of commercial indicators for the assessment of university research. Our empirical evidence shows that knowledge exchange can be detected with this approach, and, given some additions in the tools selection and adaption, it has the potential to become a supplementary method for the research community.

Abstract 579: Magnetic resonance imaging of the breast and radiomics analysis for an improved early prediction of the response to neoadjuvant chemotherapy in breast cancer patients

Abstract Background and Aim In patients undergoing neoadjuvant chemotherapy for breast cancer the achievement of a pathological complete response (pCR) is associated with a significantly improved disease-free and overall survival. Therefore, accurate means to predict treatment response as early as possible are desirable to identify women who don't benefit from this cytotoxic therapy. Several studies have demonstrated that dynamic contrast-enhanced MRI is the most sensitive method for the assessment and prediction of treatment response. In the past decade, the field of medical image analysis has grown exponentially, with an increased number of pattern recognition tools and an increase in data set sizes. These advances have facilitated the development of processes for high-throughput extraction of quantitative features that result in the conversion of images into mineable data and the subsequent analysis of these data for decision support. This emerging field in medical research is termed radiomics. The aim of this study was to assess radiomics with dynamic contrast-enhanced (DCE) MRI for the early prediction of pathological complete response (pCR) in breast cancer patients undergoing neoadjuvant chemotherapy. Methods and materials In this IRB approved prospective study 41 women with breast cancer scheduled for NAC were included and underwent MRI of the breast at 3T with DCE and T2-weighted imaging prior to and after two cycles of NAC. For each lesion a total of 14 features were extracted ranging from morphological and kinetic MRI ADC parameters. A recursive feature elimination method along with five different classifiers was performed including: linear support vector machine (SVM), linear discriminant analysis (LDA), logistic regression (LR), random forest (RF), and stochastic gradient descent (SGD) was employed to rank the features. Histopathology using the Residual Cancer Burden (RCB) score and class calculated from post-treatment surgical specimen and patient outcomes were used as the standard of reference. Results Classification accuracy was assessed for pCR as defined by the RCB score, metastases and disease specific death. Radiomics analysis of MRI data achieved AUCs for RCB score (AUC 0.85), metastases (AUC 0.87) based on RF and death (AUC 0.92) based on SVM. The most relevant parameters for prediction of RCB score were mass internal enhancement characteristics, shape and margins with DCE, for metastasis peri-tumoral edema on T2-weighted imaging, mass margins and internal enhancement characteristics, and for death high signal intensity on T2-weighted imaging, mass margins, and internal enhancement characteristics. Conclusion Radiomics with MRI of the breast using DCE and T2-weighted imaging enables prediction of response to NAC with high accuracy and thus can provide predictive information to guide treatment decisions. Citation Format: Katja Pinker-Domenig, Amirhessam Tahmassebi, Georg Wengert, Thomas H. Helbich, Zsuzsanna Bago-Horvath, Elizabeth A. Morris, Anke Meyer-Baese. Magnetic resonance imaging of the breast and radiomics analysis for an improved early prediction of the response to neoadjuvant chemotherapy in breast cancer patients [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2018; 2018 Apr 14-18; Chicago, IL. Philadelphia (PA): AACR; Cancer Res 2018;78(13 Suppl):Abstract nr 579.

A Simple Line Clustering Method for Spatial Analysis with Origin-Destination Data and Its Application to Bike-Sharing Movement Data

Clustering methods are popular tools for pattern recognition in spatial databases. Existing clustering methods have mainly focused on the matching and clustering of complex trajectories. Few studies have paid attention to clustering origin-destination (OD) trips and discovering strong spatial linkages via OD lines, which is useful in many areas such as transportation, urban planning, and migration studies. In this paper, we present a new Simple Line Clustering Method (SLCM) that was designed to discover the strongest spatial linkage by searching for neighboring lines for every OD trip within a certain radius. This method adopts entropy theory and the probability distribution function for parameter selection to ensure significant clustering results. We demonstrate this method using bike-sharing location data in a metropolitan city. Results show that (1) the SLCM was significantly effective in discovering clusters at different scales, (2) results with the SLCM analysis confirmed known structures and discovered unknown structures, and (3) this approach can also be applied to other OD data to facilitate pattern extraction and structure understanding.

Welding Defect Classification from Simulated Ultrasonic Signals

Nondestructive testing is widely used to detect and to size up discontinuities embedded in a material. Among the several ultrasonic techniques, time of flight diffraction (TOFD) combines high speed inspection, high sizing reliability and low rate of incorrect results. However, the classification of defects through ultrasound signals acquired by the TOFD technique depends heavily on the knowledge and experience of the operator and thus, this classification is still frequently questioned. Besides, this task requires long processing time due to the large amount of data to be analyzed. Nevertheless, computational tools for pattern recognition can be employed to analyze a high amount of data with large efficiency. In the present work, simulation of ultrasound propagation in two-dimensional media containing, each one, different kinds of modeled discontinuities which mimic defects in welded joints were performed. Clustering (k-means) and classification (principal component analysis and k-nearest neighbors) algorithms were employed to associate each simulated ultrasound signal with its corresponding modeled defects. The results for each method were analyzed, discussed and compared. The results are very promising.

Using pattern recognition entropy to select mass chromatograms to prepare total ion current chromatograms from raw liquid chromatography–mass spectrometry data

The total ion current chromatogram (TICC) obtained by liquid–chromatography-mass spectrometry (LC–MS) is often extremely complex and ‘noisy’ in appearance, particularly when an electrospray ionization source is used. Accordingly, meaningful qualitative and quantitative information can be obtained in LC–MS by data mining processes. Here, one or more higher-quality mass chromatograms can be identified/extracted/isolated and combined to form a TICC, wherein much of the background mass noise is eliminated, and quantitative data for chromatographic peaks can be obtained. Pattern Recognition Entropy (PRE) is a new application of Shannon’s statistical concept of entropy. PRE is both a pattern recognition tool and a summary statistic that can be used to identify information-containing mass chromatograms, where higher quality data (higher signal-to-noise mass chromatograms) usually have lower PRE values. Reduced TICCs are obtained by first calculating the PRE values of the component mass chromatograms. A plot of PRE value vs. m/z for the mass chromatograms is then generated, and the resulting band of PRE values is fit to a piecewise spline polynomial. The distribution of the differences between the individual PRE values and the spline fit is then used to select ‘good’ mass chromatograms. For the data set considered herein, best results were obtained with a threshold of 0.5 standard deviations below the average value (value of the spline). PRE reduces the number of component mass chromatograms significantly (by an order of magnitude) and at the same time preserves most of the chemical information that is collectively in them. It can also distinguish between mass chromatograms of chemically similar species. PRE is arguably a less computationally intensive alternative to the widely used CODA algorithm for variable reduction. It produces reduced TICCs of comparable if not higher quality, and it requires only a single user input for variable selection. Reduced TICCs generated by PRE can be smoothed to further improve their signal-to-noise ratios.

Characterization of hidden rules linking symptoms and selection of acupoint using an artificial neural network model.

Comprehension of the medical diagnoses of doctors and treatment of diseases is important to understand the underlying principle in selecting appropriate acupoints. The pattern recognition process that pertains to symptoms and diseases and informs acupuncture treatment in a clinical setting was explored. A total of 232 clinical records were collected using a Charting Language program. The relationship between symptom information and selected acupoints was trained using an artificial neural network (ANN). A total of 11 hidden nodes with the highest average precision score were selected through a tenfold cross-validation. Our ANN model could predict the selected acupoints based on symptom and disease information with an average precision score of 0.865 (precision, 0.911; recall, 0.811). This model is a useful tool for diagnostic classification or pattern recognition and for the prediction and modeling of acupuncture treatment based on clinical data obtained in a real-world setting. The relationship between symptoms and selected acupoints could be systematically characterized through knowledge discovery processes, such as pattern identification.

Segmentación de mallas 3d de edificios históricos para levantamiento arquitectónico

<p>Advances in three-dimensional (3D) acquisition systems have introduced this technology to more fields of study, such as archaeology or architecture. In the architectural field, scanning a building is one of the first possible steps from which a 3D model can be obtained and can be later used for visualisation and/or feature analysis, thanks to computer-based pattern recognition tools. The automation of these tools allows for temporal savings and has become a strong aid for professionals, so that more and more methods are developed with this objective. In this article, a method for 3D mesh segmentation focused on the representation of historic buildings is proposed. This type of buildings is characterised by having singularities and features in façades, such as doors or windows. The main objective is to recognise these features, understanding them as those parts of the model that differ from the main structure of the building. The idea is to use a recognition algorithm for planar faces that allows users to create a graph showing the connectivity between them, therefore allowing the reflection of the shape of the 3Dmodel. At a later step, this graph is matched against some pre-defined graphs that represent the patterns to look for. Each coincidence between both graphs indicate the position of one of the characteristics sought. The developed method has proved to be effective for feature detection and suitable for inclusion in architectural surveying applications.</p>

Chitosan-templated Pt nanocatalyst loaded mesoporous SnO2 nanofibers: a superior chemiresistor toward acetone molecules.

In this work, we introduce a chitosan-Pt complex (CS-Pt) as an effective template for catalytic Pt sensitization and creation of abundant mesopores in SnO2 nanofibers (NFs). The Pt particles encapsulated by the CS exhibit ultrasmall size (∼2.6 nm) and high dispersion characteristics due to repulsion between CS molecules. By combining CS-Pt with electrospinning, mesoporous SnO2 NFs uniformly functionalized with the Pt catalyst (CS-Pt@SnO2 NFs) are synthesized. Particularly, numerous mesopores with diameters of ∼20 nm form through the decomposition of CS, while a small SnO2 grain size (14.32 nm) is achieved by the pinning effect of CS. It is observed that CS-Pt@SnO2 NFs exhibit outstanding response (Rair/Rgas = 141.92 at 5 ppm), excellent selectivity, stability, and fast response (12 s)/recovery (44 s) speed toward 1 ppm of acetone at 350 °C and high humidity (90% RH). In addition, by applying an exponential fitting tool to experimental response values toward 0.1-5 ppm of acetone, it is estimated that CS-Pt@SnO2 NFs can detect 5 ppb of acetone with a notable response (Rair/Rgas = 2.9). Furthermore, the sensor array based on CS-Pt@SnO2 NFs, CS-driven SnO2 NFs, polyol-Pt loaded SnO2 NFs, and dense SnO2 NFs obviously classifies simulated diabetic breath and healthy human breath by using a pattern recognition tool. These results clearly demonstrate that mesoporous SnO2 NFs, particularly functionalized with CS-Pt templated nanocatalysts, open up a new class of sensing layers offering high sensitivity and selectivity.

Discriminating cereal and pseudocereal species using a binary system of GC/MS data: A pattern recognition approach

Various cultivars of different cereal and pseudocereal species (9 wheat, 8 barley, 1 rye, 3 oat, 2 triticale, 3 spelt, 12 corn, 3 amaranth and 9 buckwheat cultivar samples) were milled into flour, extracted using n-hexane, derivatized with trimethylsulfonium hydroxide solution, and subjected to GC? ?MS analysis. Fatty acid methyl esters and non-saponifiable compounds (phytosterols, ?-tocopherol and squalene) were identified by comparing mass spectra with the Wiley MS library. A binary system was applied in further data processing: the presence or the absence of a particular lipid component in each sample was coded with either (1) or (0). Major lipid components that were present in all analyzed flour samples were removed from further data analysis, leaving only those that represent a good pattern to differentiate the flour samples according to corresponding cereal/pseudocereal species. Pattern recognition tools (cluster analysis and principal component analysis) were applied to visualize groupings and separations among the samples. The presented approach enables the rapid differentiation of flour samples made from various cereal/pseudocereal species according to their botanical origin and gluten content, thereby, successfully avoiding exact quantitative determinations.

Study of microbial diversity in saliva and plaque samples from caries-free and caries-affected children using denaturing gradient gel electrophoresis.

Recent investigations have shown the possible involvement of bacteria other than mutans group and Lactobacilli in the etiology of caries. Molecular methods have been used to study the microbial diversity in caries-active (CA) and caries-free (CF) children. Among them, denaturing gradient gel electrophoresis (DGGE) is more popular and has been used in the present study. The aim of the present study was to investigate the difference in bacterial diversity in saliva and plaque samples from CF and CA children using DGGE. The study involved saliva and plaque samples from 56 children of which 28 were CF, 20 with CA, and 8 with white spot lesions (WSP). DNA was extracted and subjected to polymerase chain reaction amplification with universal primers. It was then run in polyacrylamide gel electrophoresis with gradients of urea and formamide and stained with SYBR green. Multiple bands were produced in each sample lane and each band represents one organism. A dendrogram was generated using Phoretix software and similarity index was calculated using a specific formula. Samples in each group formed several clusters indicating a specific pattern of the bacterial profile. Similarity coefficient was calculated based on the number of bands, intensity, and location. The diversity was less in the saliva and plaque samples of CA group as compared to those of CF and WSP groups. DGGE can be used to study distinctive bacterial profiles in healthy and caries-affected sites. DGGE can be further developed as a pattern recognition tool with which to identify specific groups of bacteria. Saliva may be used to study bacterial diversity in dental caries.