Search PubMed⌕ Search

SEARCH · Search PubMed

Results for “unsupervised clustering”

Search indexed PubMed citations on genomics, clinical trials, systematic reviews and public health. Explore titles, authors and supplied subject terms, then open the PubMed record.

Quote a phrase for an exact phrase match. Source license links do not imply unrestricted reuse.

At least 451 records · Page 25Linked to original sources
PubMed1998

Soft vector quantization and the EM algorithm.

The relation between hard c-means (HCM), fuzzy c-means (FCM), fuzzy learning vector quantization (FLVQ), soft competition scheme (SCS) of [Yair et al. (1992)] and probabilistic Gaussian mixtures (GM) have been pointed out recently by [Bezdek and Pal (1995)]. We extend this relation to their training, showing that learning rules by these models to estimate the cluster centers can be seen as approximations to the expectation-maximization (EM) method as applied to Gaussian mixtures. HCM and unsupervised, LVQ use 1-of-c type competition. In FCM and FLVQ, membership is the -2/(m-1)th power of the distance. In SCS and GM, Gaussian function is used. If the Gaussian membership function is used, the weighted within-groups sum of squared errors used as the fuzzy objective function corresponds to the maximum likelihood estimate in Gaussian mixtures with equal priors and covariances. The fuzzy clustering method named fuzzy c-means alternating optimization procedure (FCM-AO) proposed to optimize the former is then equivalent to batch EM and SCS's update rule is a variant of the online version of EM. The advantages of the probabilistic framework are: (i) we no longer have spurious spread parameters that needs fine tuning as m in fuzzy vector quantization or beta in SCS; instead we have a variance term that has a sound interpretation and that can be estimated from the sample; (ii) EM guarantees that the likelihood does not decrease, thus it converges to the nearest local optimum; (iii) EM also allows us to estimate the underlying distance norm and the cluster priors which we could not with the other approaches. We compare Gaussian mixtures trained with EM with LVQ (HCM), SCS and FLVQ on the IRIS dataset and see that it is more accurate due to its being able to take into account the covariance information. We finally note that vector quantization is generally an intermediate step before finding a final output for which supervision may be possible. Thus, instead of an uncoupled approach where an unsupervised method is used first to find the cluster parameters followed by supervised training of the mapping based on the memberships, we advocate a coupled approach where the cluster parameters and mapping are trained supervised in a coupled way. The uncoupled approach ignores the error at the outputs which may not be ideal.

Journal Article↗
PubMed2005

Color clustering and learning for image segmentation based on neural networks.

An image segmentation system is proposed for the segmentation of color image based on neural networks. In order to measure the color difference properly, image colors are represented in a modified L*u* v* color space. The segmentation system comprises unsupervised segmentation and supervised segmentation. The unsupervised segmentation is achieved by a two-level approach, i.e., color reduction and color clustering. In color reduction, image colors are projected into a small set of prototypes using self-organizing map (SOM) learning. In color clustering, simulated annealing (SA) seeks the optimal clusters from SOM prototypes. This two-level approach takes the advantages of SOM and SA, which can achieve the near-optimal segmentation with a low computational cost. The supervised segmentation involves color learning and pixel classification. In color learning, color prototype is defined to represent a spherical region in color space. A procedure of hierarchical prototype learning (HPL) is used to generate the different sizes of color prototypes from the sample of object colors. These color prototypes provide a good estimate for object colors. The image pixels are classified by the matching of color prototypes. The experimental results show that the system has the desired ability for the segmentation of color image in a variety of vision tasks.

Algorithms↗
PubMed2002

Segmentation of fat and muscle from MR images of the thigh by a possibilistic clustering algorithm.

Physical training is proved to induce changes in physical capacity and body composition. We propose in this article a fast, unsupervised and fully three-dimensional automatic method to extract muscle and fat volumes from magnetic resonance images of thighs in order to assess these changes. The technique relies on the use of a fuzzy clustering algorithm and post-processings to accurately process the body composition of thighs. Results are compared on 11 healthy voluntary elderly people with those provided on the same data by a validated method already published, and its reliability is assessed on repeated measures on three subjects. The two methods statistically agree when computing muscle and fat volumes, and clinical implications of this fully automatic method are important for medicine, physical conditioning, weight-loss programs and predictions of optimal body weight.

Algorithms↗
PubMed1994

Self-organized neural maps of human protein sequences.

We have recently described a method based on artificial neural networks to cluster protein sequences into families. The network was trained with Kohonen's unsupervised learning algorithm using, as inputs, the matrix patterns derived from the dipeptide composition of the proteins. We present here a large-scale application of that method to classify the 1,758 human protein sequences stored in the SwissProt database (release 19.0), whose lengths are greater than 50 amino acids. In the final 2-dimensional topologically ordered map of 15 x 15 neurons, proteins belonging to known families were associated with the same neuron or with neighboring ones. Also, as an attempt to reduce the time-consuming learning procedure, we compared 2 learning protocols: one of 500 epochs (100 SUN CPU-hours [CPU-h]), and another one of 30 epochs (6.7 CPU-h). A further reduction of learning-computing time, by a factor of about 3.3, with similar protein clustering results, was achieved using a matrix of 11 x 11 components to represent the sequences. Although network training is time consuming, the classification of a new protein in the final ordered map is very fast (14.6 CPU-seconds). We also show a comparison between the artificial neural network approach and conventional methods of biosequence analysis.

Algorithms↗
PubMed1993

A hybrid method to cluster protein sequences based on statistics and artificial neural networks.

We have recently proposed a method, based on artificial neural networks (ANNs) to cluster protein sequences into families according to their degree of sequence similarity. The network was trained with an unsupervised learning algorithm, using, as inputs, matrix patterns derived from the bipeptide composition of the protein sequences. We describe here some further improvements to that approach. First, we propose a statistical method to cluster a set of bipeptidic matrices into families. It consists of three stages: (i) principal component analysis, (ii) determination of the optimal number M of clusters and (iii) final classification of the bipeptidic matrices into M clusters. Using a set of 444 protein sequences, we show that the classification given by the statistical method is in agreement with biological knowledge. We also show that the resulting classification is very similar to the one previously obtained with the ANN approach. Finally, we propose a new hybrid method of the statistical and ANN approaches, in which the results of the statistical method are used to choose the number of neurons and inputs of the network. We show that a network built in this way, and fed with a few principal components of the set of bipeptidic matrices as input signals, can be trained in an extremely short computing time. The resulting topological maps do not essentially differ from the ones obtained with the initial ANN approach.

Algorithms↗
PubMed2001

S-TREE: self-organizing trees for data clustering and online vector quantization.

This paper introduces S-TREE (Self-Organizing Tree), a family of models that use unsupervised learning to construct hierarchical representations of data and online tree-structured vector quantizers. The S-TREE1 model, which features a new tree-building algorithm, can be implemented with various cost functions. An alternative implementation, S-TREE2, which uses a new double-path search procedure, is also developed. The performance of the S-TREE algorithms is illustrated with data clustering and vector quantization examples, including a Gauss-Markov source benchmark and an image compression application. S-TREE performance on these tasks is compared with the standard tree-structured vector quantizer (TSVQ) and the generalized Lloyd algorithm (GLA). The image reconstruction quality with S-TREE2 approaches that of GLA while taking less than 10% of computer time. S-TREE1 and S-TREE2 also compare favorably with the standard TSVQ in both the time needed to create the codebook and the quality of image reconstruction.

Algorithms↗
PubMed2026

Methylome Profiling of Cartilage Tumors: A Promising New Diagnostic Tool?

DNA methylation and copy number variation (CNV) profiling has emerged as a promising tool for the classification of bone and soft tissue tumors. We evaluated its utility in cartilage tumors, where distinguishing low-grade from high-grade conventional central chondrosarcomas (CSs) and atypical cartilaginous tumors (ACTs) from enchondromas (ECs) is a frequent diagnostic challenge, particularly on biopsy material. We analyzed 214 chondrogenic tumors, including ECs, ACTs, conventional CSs, dedifferentiated chondrosarcomas (DDCSs), and clear cell CSs, and determined their IDH1/2 mutation status. Unsupervised dimensionality reduction of genome-wide DNA methylation patterns revealed 4 clusters among IDH-mutant (MUT) tumors (IDH-MUT-1: mostly ECs and ACTs and some high-grade CSs; IDH-MUT-2: predominantly high-grade CSs; IDH-MUT-3: largely DDCSs; and IDH-MUT-SB: distinct skull base group with a markedly different methylation pattern) and 2 clusters among IDH-wild-type (WT) tumors (IDH-WT-1 and IDH-WT-2: both primarily high-grade CSs, with IDH-WT-2 showing higher tumor grade and more extensive CNVs). Clear cell CSs formed a separate cluster. The amount of CNVs, including loss of CDKN2A, increased with tumor grade, reflecting increased genomic instability during chondrosarcoma progression. Supervised classifiers trained separately, both on methylation and CNV data, and distinguished low-grade and high-grade cartilaginous tumors with area under the curve values of 0.87 to 0.97 and 85% to 90% accuracy. Furthermore, we tested whether DDCSs can be distinguished from metastatic carcinomas and other high-grade sarcomas of the bone. Across 246 reference samples, a supervised classifier achieved 97.2% accuracy (area under the curve, 99.8%) and correctly identified 30 of 32 DDCSs (93.8%). These results indicate that DNA methylation and CNV data analysis provide a valuable tool for distinguishing most low- and high-grade CSs, with additional utility also in differentiating DDCS from morphologic mimics.

cartilaginous tumors↗
PubMed1999

Classification of malignant and benign masses based on hybrid ART2LDA approach.

A new type of classifier combining an unsupervised and a supervised model was designed and applied to classification of malignant and benign masses on mammograms. The unsupervised model was based on an adaptive resonance theory (ART2) network which clustered the masses into a number of separate classes. The classes were divided into two types: one containing only malignant masses and the other containing a mix of malignant and benign masses. The masses from the malignant classes were classified by ART2. The masses from the mixed classes were input to a supervised linear discriminant classifier (LDA). In this way, some malignant masses were separated and classified by ART2 and the less distinguishable benign and malignant masses were classified by LDA. For the evaluation of classifier performance, 348 regions of interest (ROI's) containing biopsy proven masses (169 benign and 179 malignant) were used. Ten different partitions of training and test groups were randomly generated using an average of 73% of ROI's for training and 27% for testing. Classifier design, including feature selection and weight optimization, was performed with the training group. The test group was kept independent of the training group. The performance of the hybrid classifier was compared to that of an LDA classifier alone and a backpropagation neural network (BPN). Receiver operating characteristics (ROC) analysis was used to evaluate the accuracy of the classifiers. The average area under the ROC curve (A(z)) for the hybrid classifier was 0.81 as compared to 0.78 for the LDA and 0.80 for the BPN. The partial areas above a true positive fraction of 0.9 were 0.34, 0.27 and 0.31 for the hybrid, the LDA and the BPN classifier, respectively. These results indicate that the hybrid classifier is a promising approach for improving the accuracy of classification in CAD applications.

Biopsy↗
PubMed2005

Molecular phenotyping of human endometrium distinguishes menstrual cycle phases and underlying biological processes in normo-ovulatory women.

Histological evaluation of endometrium has been the gold standard for clinical diagnosis and management of women with endometrial disorders. However, several recent studies have questioned the accuracy and utility of such evaluation, mainly because of significant intra- and interobserver variations in histological interpretation. To examine the possibility that biochemical or molecular signatures of endometrium may prove to be more useful, we have investigated whole-genome molecular phenotyping (54,600 genes and expressed sequence tags) of this tissue sampled across the cycle in 28 normo-ovulatory women, using high-density oligonucleotide microarrays. Unsupervised principal component analysis of all samples revealed that samples self-cluster into four groups consistent with histological phenotypes of proliferative (PE), early-secretory (ESE), mid-secretory (MSE), and late-secretory (LSE) endometrium. Independent hierarchical clustering analysis revealed equivalent results, with two major dendrogram branches corresponding to PE/ESE and MSE/LSE and sub-branching into the four respective phases with heterogeneity among samples within each sub-branch. K-means clustering of genes revealed four major patterns of gene expression (high in PE, high in ESE, high in MSE, and high in LSE), and gene ontology analysis of these clusters demonstrated cycle-phase-specific biological processes and molecular functions. Six samples with ambiguous histology were identically assignable to a cycle phase by both principal component analysis and hierarchical clustering. Additionally, pairwise comparisons of relative gene expression across the cycle revealed genes/families that clearly distinguish the transitions of PE-->ESE, ESE-->MSE, and MSE-->LSE, including receptomes and signaling pathways. Select genes were validated by quantitative RT-PCR. Overall, the results demonstrate that endometrial samples obtained by two different sampling techniques (biopsy and curetting hysterectomy specimens) from subjects who are as normal as possible in a human study and including those with unknown histology, can be classified by their molecular signatures and correspond to known phases of the menstrual cycle with identical results using two independent analytical methods. Also, the results enable global identification of biological processes and molecular mechanisms that occur dynamically in the endometrium in the changing steroid hormone milieu across the menstrual cycle in normo-ovulatory women. The results underscore the potential of gene expression profiling for developing molecular diagnostics of endometrial normalcy and abnormalities and identifying molecular targets for therapeutic purposes in endometrial disorders.

Adult↗
PubMed1999

Unsupervised visual learning of three-dimensional objects using a modular network architecture.

This paper presents a modular network architecture that learns to cluster multiple views of multiple three-dimensional (3D) objects. The proposed network model is based on a mixture of non-linear autoencoders, which compete to encode multiple views of each 3D object. The main advantage of using a mixture of autoencoders is that it can capture multiple non-linear sub-spaces, rather than multiple centers for describing complex shapes of the view distributions. The unsupervised training algorithm is formulated within a maximum-likelihood estimation framework. The performance of the modular network model is evaluated through experiments using synthetic 3D wire-frame objects and gray-level images of real 3D objects. It is shown that the performance of the modular network model is superior to the performance of the conventional clustering algorithms, such as the K-means algorithm and the Gaussian mixture model.

Journal Article↗
PubMed1995

'Morphing' class filter: an interactive tool for continuous adjustment of tissue type related contrast.

The proposed class filter increases tissue type related contrast in MR images of brain. During the first phase of the filtering process tissue type classes are defined. This is done by operator intervention, or by a semiautomatic process based either on a supervised or unsupervised classifier, respectively. During the second phase a pixel intensity transform makes pixels of the same tissue class appear 'more similar' while the pixel intensities of different classes will become 'more different', in effect increasing the tissue type related contrast. For example, normal mixture cluster analysis is performed on an MR image set obtained by varying pulse sequence (PS) parameters and provides unsupervised definition of classes while taking advantage of much greater information content of the whole image set in comparison to that of a single image. The algorithm permits continuous transition ('morphing') between the original image and the tissue classification image that has been calculated from the input image set by simple sliding cursor-bar on the computer screen under the physician's control. Consequently, the resulting images do not require retraining of the physician who is already familiar with the appearance of standard MR images and they make mental integration of information from a large input image set possible and easier.

Brain↗
PubMed2006

Dissecting trait heterogeneity: a comparison of three clustering methods applied to genotypic data.

BACKGROUND: Trait heterogeneity, which exists when a trait has been defined with insufficient specificity such that it is actually two or more distinct traits, has been implicated as a confounding factor in traditional statistical genetics of complex human disease. In the absence of detailed phenotypic data collected consistently in combination with genetic data, unsupervised computational methodologies offer the potential for discovering underlying trait heterogeneity. The performance of three such methods--Bayesian Classification, Hypergraph-Based Clustering, and Fuzzy k-Modes Clustering--appropriate for categorical data were compared. Also tested was the ability of these methods to detect trait heterogeneity in the presence of locus heterogeneity and/or gene-gene interaction, which are two other complicating factors in discovering genetic models of complex human disease. To determine the efficacy of applying the Bayesian Classification method to real data, the reliability of its internal clustering metrics at finding good clusterings was evaluated using permutation testing. RESULTS: Bayesian Classification outperformed the other two methods, with the exception that the Fuzzy k-Modes Clustering performed best on the most complex genetic model. Bayesian Classification achieved excellent recovery for 75% of the datasets simulated under the simplest genetic model, while it achieved moderate recovery for 56% of datasets with a sample size of 500 or more (across all simulated models) and for 86% of datasets with 10 or fewer nonfunctional loci (across all simulated models). Neither Hypergraph Clustering nor Fuzzy k-Modes Clustering achieved good or excellent cluster recovery for a majority of datasets even under a restricted set of conditions. When using the average log of class strength as the internal clustering metric, the false positive rate was controlled very well, at three percent or less for all three significance levels (0.01, 0.05, 0.10), and the false negative rate was acceptably low (18 percent) for the least stringent significance level of 0.10. CONCLUSION: Bayesian Classification shows promise as an unsupervised computational method for dissecting trait heterogeneity in genotypic data. Its control of false positive and false negative rates lends confidence to the validity of its results. Further investigation of how different parameter settings may improve the performance of Bayesian Classification, especially under more complex genetic models, is ongoing.

Algorithms↗
PubMed2001

Textquest: document clustering of Medline abstracts for concept discovery in molecular biology.

We present an algorithm for large-scale document clustering of biological text, obtained from Medline abstracts. The algorithm is based on statistical treatment of terms, stemming, the idea of a 'go-list', unsupervised machine learning and graph layout optimization. The method is flexible and robust, controlled by a small number of parameter values. Experiments show that the resulting document clusters are meaningful as assessed by cluster-specific terms. Despite the statistical nature of the approach, with minimal semantic analysis, the terms provide a shallow description of the document corpus and support concept discovery.

Abstracting and Indexing↗
PubMed1991

Topological maps of protein sequences.

A new method based on neural networks to cluster proteins into families is described. The network is trained with the Kohonen unsupervised learning algorithm, using matrix pattern representations of the protein sequences as inputs. The components (x, y) of these 20 x 20 matrix patterns are the normalized frequencies of all pairs xy of amino acids in each sequence. We investigate the influence of different learning parameters in the final topological maps obtained with a learning set of ten proteins belonging to three established families. In all cases, except in those where the synaptic vectors remains nearly unchanged during learning, the ten proteins are correctly classified into the expected families. The classification by the trained network of mutated or incomplete sequences of the learned proteins is also analysed. The neural network gives a correct classification for a sequence mutated in 21.5% +/- 7% of its amino acids and for fragments representing 7.5% +/- 3% of the original sequence. Similar results were obtained with a learning set of 32 proteins belonging to 15 families. These results show that a neural network can be trained following the Kohonen algorithm to obtain topological maps of protein sequences, where related proteins are finally associated to the same winner neuron or to neighboring ones, and that the trained network can be applied to rapidly classify new sequences. This approach opens new possibilities to find rapid and efficient algorithms to organize and search for homologies in the whole protein database.

Amino Acid Sequence↗
PubMed2002

Fuzzy clustering-based segmented attenuation correction in whole-body PET imaging.

Segmented attenuation correction is now a widely accepted technique to reduce noise propagation from transmission scanning in positron emission tomography (PET). In this paper, we present a new method for segmenting transmission images in whole-body scanning. This reduces the noise in the correction maps while still correcting for differing attenuation coefficients of specific tissues. Based on the fuzzy C-means (FCM) algorithm, the method segments the PET transmission images into a given number of clusters to extract specific areas of differing attenuation such as air, the lungs and soft tissue, preceded by a median filtering procedure. The reconstructed transmission image voxels are, therefore, segmented into populations of uniform attenuation based on knowledge of the human anatomy. The clustering procedure starts with an overspecified number of clusters followed by a merging process to group clusters with similar properties (redundant clusters) and removal of some undesired substructures using anatomical knowledge. The method is unsupervised, adaptive and allows the classification of both pre- or post-injection transmission images obtained using either coincident 68Ge or single-photon 137Cs sources into main tissue components in terms of attenuation coefficients. A high-quality transmission image of the scanner bed is obtained from a high statistics scan and added to the transmission image. The segmented transmission images are then forward projected to generate attenuation correction factors to be used for the reconstruction of the corresponding emission scan. The technique has been tested on a chest phantom simulating the lungs, heart cavity and the spine, the Rando-Alderson phantom, and whole-body clinical PET studies showing a remarkable improvement in image quality, a clear reduction of noise propagation from transmission into emission data allowing for reduction of transmission scan duration. There was very good correlation (R2 = 0.96) between maximum standardized uptake values (SUVs) in lung nodules measured on images reconstructed with measured and segmented attenuation correction with a statistically significant decrease in SUV (17.03% +/- 8.4%, P < 0.01) on the latter images, whereas no proof of statistically significant differences on the average SUVs was observed. Finally, the potential of the FCM algorithm as a segmentation method and its limitations as well as other prospective applications of the technique are discussed.

Algorithms↗
PubMed2005

Global gene expression as a function of germline genetic variation.

Common, functional, germline genetic polymorphisms have been associated with clinical cancer outcomes. Little attention has been paid to the potential phenotypic consequences of germline genetic variation on downstream genes. We determined the germline status of 16 well-characterized functional polymorphisms in 126 children with newly diagnosed acute lymphoblastic leukemia (ALL). We assessed whether global gene expression profiles of diagnostic ALL blasts from the same patients differed by these germline polymorphic genotypes. Gene expression values were adjusted for ALL-subtype-specific patterns. Of the 16 loci, only the UGT1A1 promoter repeat polymorphism [A(TA)nTAA] (UGT1A1*28) and GSTM1 deletion were significant predictors of global gene expression in a supervised approach, which divided patients based on their germline genotypes [UGT1A1: 124 probe sets, false discovery rate (FDR)=13%, P< or =0.0031; GSTM1: 112 probe sets, FDR=42.5%, P< or =0.0084]. Genes whose expression distinguished the UGT1A1 (TA) 7/7 genotype from the other UGT1A1 genotypes included HDAC1, RELA and SLC2A1; those that distinguished the GSTM1 null genotype from non-null genotype included NBS1 and PRKR. In an unsupervised approach, the gene expression profiles using the entire array delineated two major clusters of patients. The only germline genotype frequency that differed between the two clusters was UGT1A1 (P=0.002; Fisher's exact test). Although their expression is limited to specific tissues, both GSTM1 and UGT1A1 are involved in the conjugation (and thus transport, excretion and lipophilicity) of a broad range of endobiotics and xenobiotics, which could plausibly have consequences for gene expression in different tissues.

Child↗
PubMed2001

[Classification study on the marshland in endemic areas of Schistosoma japonicum using satellite TM image data].

OBJECTIVE: To create a land cover map of the marshland region of Schistosoma Japonicum using satellite TM data from Chayegang, Jiangxi Province. METHODS: Satellite data of three TM bands, namely TM3, TM4 and TM5, were selected to be used in classification study with Principal Components Analysis. Then TM satellite images were applied to false color composite and unsupervised classification with computer software to manage the TM data. Finally the resulting clusters were identified by undertaking site visits and a systematic sampling snail survey was also carried out in the study area in April, 2000. RESULTS: Three land cover classes were generated, including class 1, the carex zone that is both the dominant vegetation and main snail habitats of the study area. Based on the snail survey in the spring of 2000, the live snail density and infectious snail density of the carex zone were 2.51/0.11 m2 and 0.006 9/0.11 m2, respectively. The infection rate of the snail was 0.28%. No snail was found in land cover class 3, the mixed vegetation, because the altitude of this area is either above 17 m or under 14 m and therefore the vegetation in this area is unsuitable for snails to survive. CONCLUSION: Remote sensing can contribute the study on the distribution of snail habitats and become a promising epidemiological study method for schistosomiasis and other ecological diseases.

Animals↗
PubMed1993

Protein classification using neural networks.

We have recently described a method based on Artificial Neural Networks to cluster protein sequences into families. The network was trained with Kohonen's unsupervised-learning algorithm using, as inputs, matrix patterns derived from the bipeptide composition of the proteins. We show here the application of that method to classify 1758 protein sequences, using as inputs a limited number of principal components of the bipeptidic matrices. As a result of training, the network self-organized the activation of its neurons into a topologically ordered map, in which proteins belonging to a known family (immunoglobulins, actins, interferons, myosins, HLA histocompatibility antigens, hemoglobins, etc.) were usually associated with the same neuron or with neighboring ones. Once the topological map has been obtained, the classification of new sequences is very fast.

Algorithms↗