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.

SourceResearch connectionFrom yearThrough year

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

At least 163 records · Page 9Linked to original sources

PubMed2026

Development and Validation of a Prognostic Signature Based on Transcription Factors Associated with Endoplasmic Reticulum Stress in Pancreatic Adenocarcinoma.

BACKGROUND: Endoplasmic reticulum stress (ER stress) plays a crucial role in influencing the malignant behaviors of various tumors. Targeting the expression or degradation of transcription factors (TFs) offers a promising avenue for cancer treatment. However, a detailed understanding of how ER stress affects TF function and their interactions remains limited. This study aims to develop a prognostic model and identify TFs associated with ER stress in pancreatic ductal adenocarcinoma (PDAC). METHODS: We obtained gene expression profiles and corresponding clinical data from The Cancer Genome Atlas (TCGA). To develop a prognostic signature, we performed several analyses, including unsupervised clustering, enrichment analysis, immune infiltration assessment, as well as univariate, LASSO, and multivariate Cox regression analyses. Four transcription factors-STAT1, IRF6, NRF1, and RXRA-were incorporated into a risk model, which was subsequently validated using the GSE dataset. Additionally, we examined IRF6 through quantitative PCR, western blotting, flow cytometry, and immunohistochemistry in vitro using pancreatic cancer cell lines and a tissue microarray. RESULTS: The high-risk group identified by the model exhibited significant associations with immune cell infiltration and poorer survival outcomes, though there was no significant correlation with tumor purity (p = 0.19). Furthermore, IRF6 downregulation in vitro was found to inhibit pancreatic cancer cell proliferation and promote apoptosis. IRF6 depletion also increased the expression of key molecules involved in ER stress at both the transcriptional and translational levels. Immunohistochemical analysis revealed marked differences in IRF6 expression between tumor and adjacent non-tumor tissues (59.29±29.88 vs. 95.22±40.80, p<0.001). CONCLUSION: This study provides evidence that the constructed risk model can effectively predict prognosis in PDAC patients. Transcription factors related to ER stress, such as IRF6, show promise as both prognostic biomarkers and potential therapeutic targets for PDAC.

Humans↗

PubMed2024

Developing a machine learning-based prognosis and immunotherapeutic response signature in colorectal cancer: insights from ferroptosis, fatty acid dynamics, and the tumor microenvironment.

INSTRUCTION: Colorectal cancer (CRC) poses a challenge to public health and is characterized by a high incidence rate. This study explored the relationship between ferroptosis and fatty acid metabolism in the tumor microenvironment (TME) of patients with CRC to identify how these interactions impact the prognosis and effectiveness of immunotherapy, focusing on patient outcomes and the potential for predicting treatment response. METHODS: Using datasets from multiple cohorts, including The Cancer Genome Atlas (TCGA) and Gene Expression Omnibus (GEO), we conducted an in-depth multi-omics study to uncover the relationship between ferroptosis regulators and fatty acid metabolism in CRC. Through unsupervised clustering, we discovered unique patterns that link ferroptosis and fatty acid metabolism, and further investigated them in the context of immune cell infiltration and pathway analysis. We developed the FeFAMscore, a prognostic model created using a combination of machine learning algorithms, and assessed its predictive power for patient outcomes and responsiveness to treatment. The FeFAMscore signature expression level was confirmed using RT-PCR, and ACAA2 progression in cancer was further verified. RESULTS: This study revealed significant correlations between ferroptosis regulators and fatty acid metabolism-related genes with respect to tumor progression. Three distinct patient clusters with varied prognoses and immune cell infiltration were identified. The FeFAMscore demonstrated superior prognostic accuracy over existing models, with a C-index of 0.689 in the training cohort and values ranging from 0.648 to 0.720 in four independent validation cohorts. It also responses to immunotherapy and chemotherapy, indicating a sensitive response of special therapies (e.g., anti-PD-1, anti-CTLA4, osimertinib) in high FeFAMscore patients. CONCLUSION: Ferroptosis regulators and fatty acid metabolism-related genes not only enhance immune activation, but also contribute to immune escape. Thus, the FeFAMscore, a novel prognostic tool, is promising for predicting both the prognosis and efficacy of immunotherapeutic strategies in patients with CRC.

Ferroptosis↗

PubMed2026

A murine model of sepsis induces age- and sex-specific chromatin remodeling in myeloid-derived suppressor cells.

INTRODUCTION: Sepsis survivors frequently develop long-term immune dysfunction, but the epigenetic mechanisms underlying persistent myeloid suppression remain unclear. Myeloid-derived suppressor cells (MDSCs), whose function is shaped by host age and sex, are key contributors to post-sepsis immune dysregulation. METHODS: Here, we present a high-resolution epigenetic map targeting gene promoters of MDSCs after sepsis and daily chronic stress using MAPit-FENGC, a single-molecule assay that simultaneously profiles DNA methylation and chromatin accessibility. In a clinically relevant murine model, including young and older adult male and female mice, splenic MDSCs were isolated for MAPit-FENGC and single-cell RNA sequencing. RESULTS: Unsupervised clustering identified nine promoter classes reflecting chromatin dynamics: age- and sex-dependent sepsis-induced opening (Classes 1-4), persistent closure with varying levels of DNA methylation (Classes 5-7), and constitutive openness post-sepsis (Classes 8, 9). Transcriptomic profiling corroborated these promoter states, linking accessibility with gene expression. CONCLUSIONS: These findings define promoter-level epigenetic classes across a targeted locus panel in splenic CD11b+Gr1+ cells within this murine sepsis model and generate mechanistic hypotheses regarding age- and sex-associated chromatin states.

Animals↗

PubMed2026

Efferocytosis related KCTD12 is a clinico-immune target in lung adenocarcinoma.

BACKGROUND: Efferocytosis, the clearance of apoptotic cells by phagocytes, contributes to immune homeostasis but may also promote tumor immune tolerance. However, its transcriptional landscape and clinical relevance in lung adenocarcinoma (LUAD) remain incompletely understood. METHODS: We systematically analyzed efferocytosis-associated genes across TCGA and multiple GEO datasets to classify LUAD subtypes and construct a prognostic risk model. The prognostic and immunological relevance of this model was validated in four independent cohorts and further assessed through immune infiltration, genomic, and immunotherapy datasets. Functional and pharmacogenomic analyses were performed to identify potential therapeutic vulnerabilities, and the efferocytosis-associated signature gene KCTD12 was subsequently validated in vitro. RESULTS: Unsupervised clustering identified two efferocytosis-based LUAD subtypes with distinct prognostic and immune-metabolic characteristics. The derived risk model robustly predicted overall survival across validation cohorts. Among the model genes, KCTD12 emerged as an efferocytosis-associated candidate linked to an immune-active tumor microenvironment. Across the analyzed single-cell, spatial transcriptomic, and immunotherapy-treated cohorts, higher KCTD12 expression was associated with enhanced cytotoxic T-cell activity and more favorable treatment outcomes. Functional experiments confirmed that KCTD12 suppresses tumor cell proliferation, reduces colony formation, and enhances OT-1 CD8+ T-cell activation and cytotoxicity. CONCLUSIONS: Our study identifies an efferocytosis-associated transcriptional program linked to immune heterogeneity and prognosis in LUAD. The efferocytosis-related risk signature provides a framework for prognostic and immune stratification, while KCTD12 represents a candidate biomarker associated with immune activation and clinical outcomes in immunotherapy-treated cohorts. Its treatment-specific predictive value requires prospective validation in appropriately controlled studies.

KCTD12↗

PubMed2026

CCT2 defines a highly cisplatin-resistant and poor-prognosis subtype of lung adenocarcinoma.

Cisplatin-based chemotherapy is a standard treatment for lung adenocarcinoma (LUAD), yet acquired cisplatin resistance remains a marked cause of treatment failure. The molecular mechanisms driving cisplatin resistance in LUAD have not been fully elucidated. The present study integrated bulk transcriptomic data, genomic mutation profiles and single-cell RNA sequencing data to systematically investigate cisplatin resistance in LUAD. Resistance-associated genes were identified through differential expression, survival analysis and database integration. Unsupervised clustering was used to define cisplatin resistance-associated subtypes. Functional characteristics were explored using pathway enrichment, immune infiltration, tumor mutation burden and weighted gene co-expression network analysis. A machine learning framework incorporating 101 algorithms was applied to identify key genes and construct a prognostic model. Single-cell analyses and in vitro experiments were performed to validate the biological role of the core gene. Molecular docking and molecular dynamics simulations were conducted to identify potential therapeutic compounds. A total of two molecular subtypes with distinct cisplatin resistance levels and prognostic outcomes were identified. The high-resistance subtype exhibited enhanced cell cycle activity, DNA repair signaling and immune heterogeneity. Machine learning analysis revealed a five-gene signature, with chaperonin-containing TCP1 subunit 2 (CCT2) emerging as a key regulator of cisplatin resistance. Single-cell analyses showed that CCT2 was predominantly enriched in resistant epithelial cell subpopulations. Functional experiments demonstrated that CCT2 knockdown significantly inhibited cell proliferation and enhanced cisplatin sensitivity in LUAD cell lines. A number of candidate compounds targeting CCT2 exhibited stable binding in silico. The present findings identified CCT2 as a key mediator of cisplatin resistance in LUAD and provided potential therapeutic strategies to overcome chemotherapy resistance.

chaperonin-containing TCP-1 subunit 2↗

PubMed2002

Primer on medical genomics. Part III: Microarray experiments and data analysis.

Genomics has been defined as the comprehensive study of whole sets of genes, gene products, and their interactions as opposed to the study of single genes or proteins. Microarray technology is one of many novel tools that are allowing global and high-throughput analysis of genes and gene products. In addition to an introduction on underlying principles, the current review focuses on the use of both complementary DNA and oligodeoxynucleotide microarrays in gene expression analysis. Genome-wide experiments generate a massive amount of data points that require systematic methods of analysis to extract biologically useful information. Accordingly, the current educational communication discusses different methods of data analysis, including supervised and unsupervised clustering algorithms. Illustrative clinical examples show clinical applications, including (1) identification of candidate genes or pathological pathways (ie, elucidation of pathogenesis); (2) identification of "new" molecular classes of diseases that may be relevant in disease reclassification, prognostication, and treatment selection (ie, class discovery); and (3) use of expression profiles of known disease classes to predict diagnosis and classification of unknown samples (ie, class prediction). The current review should serve as an introduction to the subject for clinician investigators, physicians and medical scientists in training, practicing clinicians, and other students of medicine.

Breast Neoplasms↗

PubMed2002

The effect of contrast and luminance on mfERG responses in a monkey model of glaucoma.

PURPOSE: To evaluate the effect of contrast and luminance attenuation on the multifocal electroretinogram (mfERG) responses of normal and glaucomatous eyes of cynomolgus monkeys. METHODS: Nine adult male cynomolgus monkeys with unilateral experimentally induced glaucoma were used. Hypertension-induced damage was confirmed by tomography of the optic disc. mfERGs were recorded with five different stimulus contrasts and/or luminance settings. The first-order and the first slice of second-order responses were analyzed. RESULTS: Waveforms of normal and glaucomatous eyes differed in shape and amplitude. Second-order responses contributed to first-order responses of the signals in the normal eyes, but made a negligible contribution to the signals in the glaucomatous eyes. Contrast and luminance attenuation affected both first- and second-order responses. The differences between signals in normal and glaucomatous eyes were sufficiently large for an unsupervised clustering algorithm to achieve accurate segregation. CONCLUSIONS: The observations in this study indicate that outer and inner retinal generators participate in first-order mfERG responses and that both inner and outer retinal contributors respond to contrast and luminance changes in stimulus. The hypertension-induced changes in the mfERG furthermore suggest damage to both inner and outer retina.

Animals↗

PubMed2003

Profiling of genes differentially expressed between fetal liver and postnatal liver using high-density oligonucleotide DNA array.

The liver is an essential organ in humans not only for the production and storage of energy but also for detoxification of chemical compounds, but knowledge about changes in the gene expression profile in the human liver during the prenatal and postnatal periods is limited. Profiling of genes differentially expressed between the fetal liver (FL) and the postnatal liver (PNL) is one of the methods to investigate candidates affecting the difference in biological characteristics between FL and PNL. To identify genes differentially expressed between FL and PNL (childhood and adult liver), we analyzed the gene expression profiles across 9 FL and 14 PNL samples using a high-density oligonucleotide DNA array. Using Mann-Whitney U test followed by k-nearest-neighbors (supervised learning method) and hierarchical clustering (unsupervised learning method) algorithms, we found 33 genes clearly discriminating between the FL group and PNL group. The functional classification of the 33 genes identified was related to several kinds of biological pathways, regulating the cell cycle (PCNA, CDC7L1, CCND3, YWHA1, PKMYT1), DNA replication and repair (RFC4, RECQ2, PCNA, NAP1L1), cell growth (IGF2, IGFBP2, PRSS11), hormonal signals (AR, SRD5A1, NR1I3), and cellular metabolism (E2-EPF, WWP1, CYP2C9, CYP2E1, CYP2A6, CYP2A7, CYP2A13, CYP4F2, CYP3A4, DDT). The results presented herein provide evidence of a differential expression profile of genes regulating the cell cycle, DNA replication and repair, cell growth, regulation of hormonal signals, and cellular metabolism, between FL and PNL in humans. The 33 genes identified in this study are suggested to be useful markers clearly discriminating between FL and PNL using the gene expression profile.

Adult↗

PubMed2003

Differentiation of lobular versus ductal breast carcinomas by expression microarray analysis.

Invasive lobular and ductal breast tumors have distinct histologies and clinical presentation. Other than altered expression of E-cadherin, little is known about the underlying biology that distinguishes the tumor subtypes. We used cDNA microarrays to identify genes differentially expressed between lobular and ductal tumors. Unsupervised clustering of tumors failed to distinguish between the two subtypes. Prediction analysis for microarrays (PAM) was able to predict tumor type with an accuracy of 93.7%. Genes that were significantly differentially expressed between the two groups were identified by MaxT permutation analysis using t tests (20 cDNA clones and 10 unique genes), significance analysis for microarrays (33 cDNA clones and 15 genes, at an estimated false discovery rate of 2%), and PAM (31 cDNAs and 15 genes). There were 8 genes identified by all three of these related methods (E-cadherin, survivin, cathepsin B, TPI1, SPRY1, SCYA14, TFAP2B, and thrombospondin 4), and an additional 3 that were identified by significance analysis for microarrays and PAM (osteopontin, HLA-G, and CHC1). To validate the differential expression of these genes, 7 of them were tested by real-time quantitative PCR, which verified that they were differentially expressed in lobular versus ductal tumors. In conclusion, specific changes in gene expression distinguish lobular from ductal breast carcinomas. These genes may be important in understanding the basis of phenotypic differences among breast cancers.

Breast Neoplasms↗

PubMed2003

Text mining of DNA sequence homology searches.

Primary tasks in analysis and annotation of expressed sequence tag (EST) datasets are to identify similarity among sequences by unsupervised clustering and assign putative function based on BLAST homology searches. We investigated the usefulness of text mining as a simple approach for further higher-level clustering of EST datasets using IBM Intelligent Miner for Text v2.3 tools. Agglomerative and k-means clustering tools were used to cluster BLASTx homology search documents from two onion EST datasets and optimised by pre-processing and pruning. Subjective evaluation confirmed that these tools provided biologically useful and complementary views of the two libraries, provided new insights into their composition and revealed clusters previously identified by human experts. We compared BLASTx textual clusters for two gene families with their DNA sequence-based clusters and confirmed that these shared similar morphology.

Abstracting and Indexing↗

PubMed2005

Gene expression profiles in prostate cancer: association with patient subgroups and tumour differentiation.

Prostate carcinoma is the most common cancer of western men and is a markedly heterogeneous disease. The aim of this study was to identify signatures of differentially expressed genes in prostate cancer using DNA microarray technology, evaluating expression profiles in matched pairs of benign and malignant tissue. Samples were collected from 33 radical prostatectomies, and 52 specimens were included, representing 29 histologically verified primary tumours, 19 paired samples of malignant and benign tissue, and 4 non-paired benign tissue samples. Microarray analysis was performed using an expanded sequence verified set of 40,000 human cDNA clones, revealing several genes with significant differences between malignant and benign tissue, including recently reported genes like alpha-methylacyl-CoA racemase (AMACR) and hepsin, as well as genes relevant for tumour development and progression. Leave out cross validation (LOCV) test correctly predicted tumour or benign tissue in 47 (90.3%) out of 52 cases, significantly better than cross validation tests using randomly permuted tissue labels. Unsupervised clustering analysis revealed 3 distinct patient clusters significantly associated with Gleason score, and high grade tumours (Gleason score >/=7) accumulated in cluster 1 (C1). Gene expression profiles correctly predicted 100% of tumour samples segregating to C1, as also validated by LOCV. Gene expression profiles were analysed in filtered and floored datasets with similar results, and a pair-wise design was also tested. Gene expression profiles provided tumour clusters linked to differentiation, and revealed novel markers relevant for molecular classification, grading and therapy of prostate cancer.

Cluster Analysis↗

PubMed2005

Peritoneal and subperitoneal stroma may facilitate regional spread of ovarian cancer.

PURPOSE: Epithelial ovarian cancer (EOC) is characterized by early peritoneal involvement ultimately contributing to morbidity and mortality. To study the role of the peritoneum in fostering tumor invasion, we analyzed differences between the transcriptional repertoires of peritoneal tissue lacking detectable cancer in patients with EOC versus benign gynecologic disease. EXPERIMENTAL DESIGN: Specimens were collected at laparotomy from patients with benign disease (b) or malignant (m) ovarian pathology and comprised primary ovarian tumors, paired bilateral specimens from adjacent peritoneum and attached stroma (PE), subjacent stroma (ST), peritoneal washes, ascites, and peripheral blood mononuclear cells. Specimens were immediately frozen. RNA was amplified by in vitro transcription and cohybridized with reference RNA to a custom-made 17.5k cDNA microarray. RESULTS: Principal component analysis and unsupervised clustering did not segregate specimens from patients with benign or malignant pathology. Class comparison identified differences between benign and malignant PE and ST specimens deemed significant by permutation test (P = 0.027 and 0.012, respectively). A two-tailed Student's t test identified 402 (bPE versus mPE) and 663 (mST versus bST) genes differentially expressed at a significance level of P2 < or = 0.005 when all available paired samples from each patient were analyzed. The same comparison using one sample per patient reduced the pool of differentially expressed genes but retained permutation test significance for bST versus mST (P = 0.031) and borderline significance for bPE versus mPE (P = 0.056) differences. CONCLUSIONS: The presence of EOC may foster peritoneal implantation and growth of cancer cells by inducing factors that may represent molecular targets for disease control.

Computational Biology↗

PubMed2006

Alteration of hTERT full-length variant expression level showed different gene expression profiles and genomic copy number changes in breast cancer.

We analyzed hTERT splicing patterns with respect to telomerase activity in breast cancer. Using a cDNA microarray in 22 cell lines, we observed the difference in expression profiling based on the different levels of full-length variant expression with 71 selected genes. Using 33 known genes that act with the telomerase complex, we performed unsupervised clustering with all cell lines, and found a clustering tendency related to the full-length variant expression level. Using array-based CGH, highly altered genomic copy number changes were found more often in MCF-7 (159 genes) than in MDA-MB-231 (109 genes) and MDA-MB-435 (49 genes), suggesting more genomic changes in MCF-7 cells. On comparing MCF-7 with MDA-MB231 and MDA-MB-435 cell lines, we identified 8 genes with different copy numbers, including dystroglycan, which is located in the p12-21.2 area of chromosome 3. In conclusion, alterations in the level of the full-length variant of hTERT showed different gene expression profiles and genomic copy number changes in breast cancer, which require further study into their cause-and-effect relationship.

Alternative Splicing↗

PubMed2006

Identification of gene expression signatures for molecular classification in human leukemia cells.

Although the methods by which leukemia is classified have been improved for effective therapies, leukemia patients occasionally exhibit diverse, sometimes unpredictable, responses to treatment. Consequently, these patients also evidence individually different clinical courses when administered with anti-leukemia drugs. In order to find new, more precise molecular markers for leukemia classification, we have analyzed the gene expression profiles from 65 diagnostic bone marrow specimens of adult patients with AML, ALL, CML or CLL by using high-throughput DNA microarrays harboring approximately 8,300 unique human genes or expression sequence tags. In the present study, we identified a group of leukemia-specific genes, which manifest gene expression profiles distinctly representative of normal bone marrow samples, as determined by a significance analysis of microarray (SAM) and GeneSpring 6.1 programs. We also determined the minimal number of genes showing a difference between acute and chronic leukemia patient groups. Furthermore, the unsupervised cluster analysis revealed a gene subset which can be used to distinguish between AML, ALL, CML and CLL patient groups, based on expression signatures. The expression levels of differentially regulated genes were verified via the principle component analysis (PCA). Our results may provide a novel set of molecular criteria for the classification of leukemia patients, and may also facilitate effects to discovery new targets, allowing for more effective treatment of leukemia patients.

Adolescent↗

PubMed2006

Gene expression profiles of small-cell lung cancers: molecular signatures of lung cancer.

To characterize the molecular mechanisms involved in the carcinogenesis and progression of small-cell lung cancer (SCLC) and identify molecules to be applied as novel diagnostic markers and/or for development of molecular-targeted drugs, we applied cDNA microarray profile analysis coupled with purification of cancer cells by laser-microbeam microdissection (LMM). Expression profiles of 32,256 genes in 15 SCLCs identified 252 genes that were commonly up-regulated and 851 transcripts that were down-regulated in SCLC cells compared with non-cancerous lung tissue cells. An unsupervised clustering algorithm applied to the expression data easily distinguished SCLC from the other major histological type of non-small cell lung cancer (NSCLC) and identified 475 genes that may represent distinct molecular features of each of the two histological types. In particular, SCLC was characterized by altered expression of genes related to neuroendocrine cell differentiation and/or growth such as ASCL1, NRCAM, and INSM1. We also identified 68 genes that were abundantly expressed both in advanced SCLCs and advanced adenocarcinomas (ADCs), both of which had been obtained from patients with extensive chemotherapy treatment. Some of them are known to be transcription factors and/or gene expression regulators such as TAF5L, TFCP2L4, PHF20, LMO4, TCF20, RFX2, and DKFZp547I048 as well as those encoding nucleotide-binding proteins such as C9orf76, EHD3, and GIMAP4. Our data provide valuable information for better understanding of lung carcinogenesis and chemoresistance.

Antineoplastic Agents↗

PubMed2006

Candidate early predictors for progression to joint damage in systemic juvenile idiopathic arthritis.

OBJECTIVE: To assess if joint damage at 2 years after diagnosis in patients with systemic juvenile idiopathic arthritis (SJIA) can be predicted by clinical or laboratory features assessed up to 3 or 6 months after diagnosis. METHODS: Medical records from 70 children were retrospectively reviewed. The primary outcome measure was presence of joint damage at 2 years after diagnosis (JD2) as defined by presence of erosions or fusion in one or more joints. Potential predictor variables for JD2 in the first 3 and 6 months after diagnosis consisted of the highest observed white blood cell count, platelet count, erythrocyte sedimentation rate, active joint count, and presence of symptomatic pulmonary or cardiac disease or macrophage activation syndrome, and treatment data. RESULTS: The outcome of interest, JD2, was identified in 15/70 patients. Classification-tree analysis identified a pair of variables (highest observed platelet count and number of active joints) measured within the first 3 months after diagnosis that together predicted progression to JD2 with an estimated sensitivity of 87%, specificity of 82%, and positive predictive value of 57%. Multivariate logistic regression analyses at 3 months found that higher quantities of joints with active arthritis and early use of methotrexate (MTX) were factors significantly associated with increased odds of progression to JD2 (active joints odds ratio = 1.08, 95% CI 1.00-1.16, p = 0.04; MTX OR = 11.85, 95% CI 1.89-74.26, p = 0.01). Unsupervised cluster analysis identified 2 major phenotypes of patients at 3 months characterized by different ages at onset, acute phase markers, active joint counts, and presence of serositis. These phenotypes differed 3-fold in proportion of subjects progressing to JD2 (p < 0.05). CONCLUSION: By 3 months after diagnosis, a clinical phenotype based on active joint count and platelet count may be prognostic of an increased risk of progression to JD2. Use of corticosteroids did not appear to change the risk of joint damage. In contrast, the presence of serositis appeared to be associated with decreased risk of joint damage.

Acute-Phase Proteins↗

PubMed2005

Unsupervised multistage image classification using hierarchical clustering with a Bayesian similarity measure.

A new multistage method using hierarchical clustering for unsupervised image classification is presented. In the first phase, the multistage method performs segmentation using a hierarchical clustering procedure which confines merging to spatially adjacent clusters and generates an image partition such that no union of any neighboring segments has homogeneous intensity values. In the second phase, the segments resulting from the first stage are classified into a small number of distinct states by a sequential merging operation. The region-merging procedure in the first phase makes use of spatial contextual information by characterizing the geophysical connectedness of a digital image structure with a Markov random field, while the second phase employs a context-free similarity measure in the clustering process. The segmentation procedure of region merging is implemented as a hierarchical clustering algorithm whereby a multiwindow approach using a pyramid-like structure is employed to increase computational efficiency while maintaining spatial connectivity in merging. From experiments with both simulated and remotely sensed data, the proposed method was determined to be quite effective for unsupervised analysis. In particular, the region-merging approach based on spatial contextual information was shown to provide more accurate classification of images with smooth spatial patterns.

Algorithms↗

PubMed2001

On the number of clusters and the fuzziness index for unsupervised FCA application to BOLD fMRI time series.

The aim of this paper is to present an exploratory data-driven strategy based on Unsupervised Fuzzy Clustering Analysis (UFCA) and its potential for fMRI data analysis in the temporal domain. The a priori definition of the number of clusters is addressed and solved using heuristics. An original validity criterion is proposed taking into account data geometry and the partition Membership Functions (MFs). From our simulations, this criterion is shown to outperform other indices used in the literature. The influence of the fuzziness index was studied using simulated activation combined with real life noise data acquired from subjects under a resting state. Receiver Operating Characteristics (ROC) methodology is implemented to assess the performance of the proposed UFCA with respect to the fuzziness index. An interval of choice around 2, a value widely used in FCA, is shown to yield the best performance.

Algorithms↗