SEARCH · Search PubMed

Results for “classifier”

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 415 records · Page 23Linked to original sources

PubMed2006

PACK: Profile Analysis using Clustering and Kurtosis to find molecular classifiers in cancer.

MOTIVATION: Elucidating the molecular taxonomy of cancers and finding biological and clinical markers from microarray experiments is problematic due to the large number of variables being measured. Feature selection methods that can identify relevant classifiers or that can remove likely false positives prior to supervised analysis are therefore desirable. RESULTS: We present a novel feature selection procedure based on a mixture model and a non-gaussianity measure of a gene's expression profile. The method can be used to find genes that define either small outlier subgroups or major subdivisions, depending on the sign of kurtosis. The method can also be used as a filtering step, prior to supervised analysis, in order to reduce the false discovery rate. We validate our methodology using six independent datasets by rediscovering major classifiers in ER negative and ER positive breast cancer and in prostate cancer. Furthermore, our method finds two novel subtypes within the basal subgroup of ER negative breast tumours, associated with apoptotic and immune response functions respectively, and with statistically different clinical outcome. AVAILABILITY: An R-function pack that implements the methods used here has been added to vabayelMix, available from (www.cran.r-project.org). CONTACT: aet21@cam.ac.uk SUPPLEMENTARY INFORMATION: Supplementary information is available at Bioinformatics online.

Algorithms↗

PubMed2006

Sample size planning for developing classifiers using high-dimensional DNA microarray data.

Many gene expression studies attempt to develop a predictor of pre-defined diagnostic or prognostic classes. If the classes are similar biologically, then the number of genes that are differentially expressed between the classes is likely to be small compared to the total number of genes measured. This motivates a two-step process for predictor development, a subset of differentially expressed genes is selected for use in the predictor and then the predictor constructed from these. Both these steps will introduce variability into the resulting classifier, so both must be incorporated in sample size estimation. We introduce a methodology for sample size determination for prediction in the context of high-dimensional data that captures variability in both steps of predictor development. The methodology is based on a parametric probability model, but permits sample size computations to be carried out in a practical manner without extensive requirements for preliminary data. We find that many prediction problems do not require a large training set of arrays for classifier development.

Computer Simulation↗

PubMed1998

Action potential classifiers: a functional comparison of template matching, principal components analysis and an artificial neural network.

Multiunit neural activity occurs often in electrophysiological studies when utilizing extracellular electrodes. In order to estimate the activity of the individual neurons each action potential in the recording must be classified to its neuron of origin. This paper compares the accuracy of two traditional methods of action potential classification--template matching and principal components--against the performance of an artificial neural network (ANN). Both traditional methods use averages of action potential shapes to form their corresponding classifiers while the artificial neural network 'learns' a nonlinear relationship between a set of prototype action potentials and assigned classes. The set of prototypic action potentials and the assigned classes is termed the training set. The training set contained action potentials from each class which exhibited the full range of amplitude variability. The ANN provided better classification results and was more robust in analysis of across-animal data sets than either of the traditional action potential classification methods.

Action Potentials↗

PubMed2005

Comparing action gestures and classifier verbs of motion: evidence from Australian Sign Language, Taiwan Sign Language, and nonsigners' gestures without speech.

Recent research into signed languages indicates that signs may share some properties with gesture, especially in the use of space in classifier constructions. A prediction of this proposal is that there will be similarities in the representation of motion events by sign-naive gesturers and by native signers of unrelated signed languages. This prediction is tested for deaf native signers of Australian Sign Language (Auslan), deaf signers of Taiwan Sign Language (TSL), and hearing nonsigners using the Verbs of Motion Production task from the Test Battery for American Sign Language (ASL) Morphology and Syntax. Results indicate that differences between the responses of nonsigners, Auslan signers, and TSL signers and the expected ASL responses are greatest with handshape units; movement and location units appear to be very similar. Although not definitive, these data are consistent with the claim that classifier constructions are blends of linguistic and gestural elements.

Adolescent↗

PubMed1987

Classifying family/household problems.

To investigate how family/household problems could be classified 186 problems recorded in 55 families by 14 Israeli family physicians were studied. Four major categories were used which included health problems in one family member affecting others in the household, relationship problems, social problems and health problems in more than one family member. Over half of the 186 problems were psychological or physical problems in the individual which affected other household members. Specific diagnostic titles from the International Classification of Health Problems in Primary Care (ICHPPC-2-Defined) were suitable to classify virtually all identified family/household problems. More problems were identified in mothers than in fathers or children and mothers were perceived as being affected most by social and relationship problems. Additional studies are suggested.

Adult↗

PubMed2001

A case study of using artificial neural networks for classifying cause of death from verbal autopsy.

BACKGROUND: Artificial neural networks (ANN) are gaining prominence as a method of classification in a wide range of disciplines. In this study ANN is applied to data from a verbal autopsy study as a means of classifying cause of death. METHODS: A simulated ANN was trained on a subset of verbal autopsy data, and the performance was tested on the remaining data. The performance of the ANN models were compared to two other classification methods (physician review and logistic regression) which have been tested on the same verbal autopsy data. RESULTS: Artificial neural network models were as accurate as or better than the other techniques in estimating the cause-specific mortality fraction (CSMF). They estimated the CSMF within 10% of true value in 8 out of 16 causes of death. Their sensitivity and specificity compared favourably with that of data-derived algorithms based on logistic regression models. CONCLUSIONS: Cross-validation is crucial in preventing the over-fitting of the ANN models to the training data. Artificial neural network models are a potentially useful technique for classifying causes of death from verbal autopsies. Large training data sets are needed to improve the performance of data-derived algorithms, in particular ANN models.

Autopsy↗

PubMed2003

Men classified as hypo- or hyperresponders to dietary cholesterol feeding exhibit differences in lipoprotein metabolism.

The purpose of this study was to evaluate the differences that occur within the plasma compartment of normolipidemic men, classified on the basis of their response to prolonged consumption of additional dietary cholesterol. Using a crossover design, 40 men aged 18-57 y were randomly allocated to an egg (640 mg/d additional dietary cholesterol) or placebo group (0 mg/d additional dietary cholesterol), for two 30-d periods, which were separated by a 3-wk washout period. Subjects were classified as hypo- [increase in plasma total cholesterol (TC) of <0.05 mmol/L for each additional 100 mg of dietary cholesterol consumed] or hyperresponders (increase in TC of > or =0.06 mmol/L for each additional 100 mg of dietary cholesterol consumed) on the basis of their plasma reaction to the additional dietary cholesterol provided. Male hyporesponders did not experience an increase in LDL cholesterol (LDL-C) or HDL cholesterol (HDL-C) during the egg period, whereas both lipoproteins were significantly (P < 0.0001 and P < 0.05, respectively) elevated in hyperresponders. Although the LDL/HDL ratio was increased in male hyperresponders after the high cholesterol period, the mean increase experienced by this population was still within National Cholesterol Education Program guidelines. Furthermore, male hyperresponders had higher lecithin cholesterol acyltransferase (P < 0.05) and cholesteryl ester transfer protein (P < 0.05) activities during the egg period, which suggests an increase in reverse cholesterol transport. These data suggest that additional dietary cholesterol does not increase the risk of developing an atherogenic lipoprotein profile in healthy men, regardless of their response classification.

Adolescent↗

PubMed2003

A streamlined three-dimensional volume estimation method accurately classifies prostate tumors by volume.

Prostate tumor volume has been suggested to be an important pathologic variable that predicts for clinical significance and outcome. However, the determination of tumor volume using standard methods such as computerized planimetry or image analysis is labor intensive. We studied whether length (L), width (W), and height (number of cross sections x sectional thickness, CST) of a tumor focus could be used to estimate prostate tumor volume. We studied 1091 tumor foci from 365 selected serially sectioned radical prostatectomy specimens. We randomly divided the specimens into evaluation (182 specimens) and validation (183 specimens) groups. After analyzing the evaluation group, we derived the formula 0.4 (slope of the regression line) x L x W x CST to estimate volume. We then tested whether our three-dimensional volume estimation formula could accurately classify tumor volume for specimens in the validation set as insignificant (</=0.5 cm3) or significant (>0.5 cm3), and also into a five-category tumor volume scheme. Our three-dimensional estimate accurately classified tumors into insignificant and significant total volume categories in 94.0% of cases and into the five-category scheme in 85.8% cases. These accuracy rates were significantly better than rates for other methods. The three-dimensional estimate is an accurate and straightforward method for assessing prostate tumor volume.

Body Weights and Measures↗

PubMed2000

Electroencephalogram approximate entropy correctly classifies the occurrence of burst suppression pattern as increasing anesthetic drug effect.

BACKGROUND: Approximate entropy, a measure of signal complexity and regularity, quantifies electroencephalogram changes during anesthesia. With increasing doses of anesthetics, burst-suppression patterns occur. Because of the high-frequency bursts, spectrally based parameters such as median electroencephalogram frequency and spectral edge frequency 95 do not decrease, incorrectly suggesting lightening of anesthesia. The authors investigated whether the approximate entropy algorithm correctly classifies the occurrence of burst suppression as deepening of anesthesia. METHODS: Eleven female patients scheduled for elective major surgery were studied. After propofol induction, anesthesia was maintained with isoflurane only. Before surgery, the end-tidal isoflurane concentration was varied between 0.6 and 1.3 minimum alveolar concentration. The raw electroencephalogram was continuously recorded and sampled at 128 Hz. Approximate entropy, electroencephalogram median frequency, spectral edge frequency 95, burst-suppression ratio, and burst-compensated spectral edge frequency 95 were calculated offline from 8-s epochs. The relation between burst-suppression ratio and approximate entropy, electroencephalogram median frequency, spectral edge frequency 95, and burst-compensated spectral edge frequency 95 was analyzed using Pearson correlation coefficient. RESULTS: Higher isoflurane concentrations were associated with higher burst-suppression ratios. Electroencephalogram median frequency (r = 0.34) and spectral edge frequency 95 (r = 0.29) increased, approximate entropy (r = -0.94) and burst-compensated spectral edge frequency 95 (r = -0.88) decreased with increasing burst-suppression ratio. CONCLUSION: Electroencephalogram approximate entropy, but not electroencephalogram median frequency or spectral edge frequency 95 without burst compensation, correctly classifies the occurrence of burst-suppression pattern as increasing anesthetic drug effect.

Adult↗

PubMed2000

Classifying choledochal cysts using hepatobiliary scintigraphy.

PURPOSE: This retrospective study was designed to classify choledochal cysts on the basis of the findings of hepatobiliary scintigraphy. METHODS: Twenty-one patients with choledochal cysts (15 female, 6 male; mean age, 20 years) proved on the findings of endoscopic retrograde cholangiopancreatography (ERCP) or surgery and histopathologic analysis were included in the study. Two nuclear medicine physicians, blinded with regard to cholangiographic and operative details, were asked to review and to classify the type of choledochal cyst seen on the hepatobiliary scan. Later, scintigraphic results were compared with ERCP and surgical findings for a reference standard. RESULTS: The findings of hepatobiliary scintigraphy correlated with ERCP and surgical findings in 18 of 21 cases (86%). Scintiscans correctly identified all type 1 cysts (12/12). The sensitivity of scintigraphy in diagnosing type 4 cysts was 66% (6 of 9 cases). It underestimated the intrahepatic extent of disease in type 4a biliary cysts (37%). CONCLUSION: This study illustrates the utility of hepatobiliary scintigraphy in diagnosing type 1 and 4 choledochal cysts.

Adult↗

PubMed1996

Pattern recognition system for focal liver lesions using "crisp" and "fuzzy" classifiers.

RATIONALE AND OBJECTIVES: To determine the diagnostic performance of an artificial intelligence system for classification of focal liver lesions, in comparison to human observers. METHODS: One hundred forty-three focal hepatic lesions were evaluated with dynamic computed tomography. The study comprised 59 hemangiomas, 24 other benign lesions (focal nodular hyperplasia, adenoma), and 60 malignant liver lesions (18 primary, 42 secondary). All lesions but the hemangiomas were histologically examined by needle biopsy. For delineation of the lesion, a region of interest was defined interactively. The pattern recognition was performed in two steps with initial extraction of textural features: training of a classifier and classification of the lesions. The accuracy of classification of hepatic lesions into three groups (hemangioma, other benign processes, malignant lesions) was tested. The results were compared with those achieved by human observers using receiver operating characteristic statistical analysis. RESULTS: The accuracy (total rate of correct diagnoses) was 90.2%. False classifications were found owing to small size, weak contrast enhancement after bolus injection, respiratory movement, and atypical morphology of the lesion. The area under the receiver operating characteristic curve was not significantly different for computer and human observers. CONCLUSIONS: The system demonstrated a diagnostic accuracy comparable to human observers. Further improvement with increasing numbers of typical computed tomographic series for training of the classifier can be expected.

Adenoma↗

PubMed2002

Diagnostic potential of targeted electrical impedance scanning in classifying suspicious breast lesions.

RATIONALE AND OBJECTIVE: To evaluate the potential of targeted electrical impedance scanning (EIS) for classifying suspicious breast lesions. METHODS: EIS was performed in full knowledge of mammographic findings and findings of clinical breast examination. One hundred seventeen patients with a total of 129 breast lesions were examined with EIS before breast biopsy (surgical excision or vacuum core biopsy). Diagnostic indexes of targeted EIS were calculated depending on major lesion characteristics. Capacitance and conductivity of all positive spots (S) and the surrounding normal breast tissue (NBT) were quantified using ROI measurements. The ratio S/NBT was calculated to compare true positive (n = 44) and false positive (n = 18) spots. RESULTS: With respect to histology, of the 129 lesions 71 were malignant and 58 lesions were benign. Overall sensitivity of targeted EIS was 62%, specificity 69%, PPV 71%, and NPV 60%. Sensitivity of EIS varied depending on the tumor size, which was between 48% (> 20 mm) and 71% (11-20 mm). Highest specificity (86%) was observed for large lesions (> 20 mm); however, the NPV was only 35% for lesions of that size. NPV was higher for nonpalpable lesions (74%) and clusters of microcalcifications (85.7%) compared with palpable lesions (39%) and solid lesions (44%). There was no statistical difference of S/NBT ratio neither for conductivity nor capacitance of true and false positive spots. Compared with true positive spots a trend of a higher conductivity ratio at 100 Hz and 200 Hz was seen for false positive spots. CONCLUSION: EIS showed mediocre overall diagnostic accuracy for classifying suspicious breast lesions. Quantitative analysis of positive EIS findings did not help to differentiate between false and true positive spots.

Biopsy↗

PubMed1996

Identifying thresholds for classifying childhood psychiatric disorder: issues and prospects.

OBJECTIVE: To evaluate empirically the implications of choosing different thresholds to classify conduct disorder and attention-deficit hyperactivity disorder for estimating prevalence, test-retest reliability of measurement, and informant (parent/teacher) agreement and for evaluating comorbidity and associated features of disorder. METHOD: Data for the study came from problem checklist assessments done by parents and teachers of children aged 6 to 16 years (N = 1,229) selected with known probability from a general population sample and from structured interviews obtained in a stratified, random subsample (n = 251). RESULTS: Estimates varied widely depending on the rationale used to set thresholds. Percent prevalence went from 0.1 to 39.2; kappa estimates of test-retest reliability went from .19 to .82. Parent-teacher agreement based on kappa went from .0 to .38. Relative odds between disorder and associated features varied twofold. CONCLUSION: Use of different rationales to set thresholds for classifying childhood psychiatric disorder in the general population has profound implications for what we learn about the epidemiology of childhood disorder.

Adolescent↗

PubMed1994

Cellular hemangioma and angioblastoma of the spine, originally classified as hemangioendothelioma. A confusing diagnosis.

STUDY DESIGN: The authors report two cases of vascular tumors of the spine, classified originally as benign and malignant hemangioendothelioma, and after revision, as cellular hemangioma and angioblastomatosis, respectively. OBJECTIVES: Problems in interpretation of the confusing term hemangioendothelioma and treatment modalities for vascular tumors of the spine are discussed. SUMMARY OF BACKGROUND DATA: Hemangioendothelioma is a confusing term and is often used to cover bewilderment at the biological behavior of a vascular tumor. Its spectrum ranges, depending the references used, from benign to malignant and can mistakenly include benign lesions like cellular hemangioma and angioblastoma (solitary and multicentric). METHODS: Of two patients with a cellular tumor of the spine, the clinicopathologic data and modes of treatment are reviewed. The relevant literature is discussed. RESULTS: In the first case, the diagnosis of benign cellular hemangioendothelioma was changed to cellular hemangioma. In the second case, the original diagnosis of malignant hemangioendothelioma with metastasis to liver and lungs was changed to angioblastomatosis, most probably benign. In both cases, a correct interpretation of the initial diagnosis or proper diagnosis would have influenced the mode of treatment. CONCLUSION: Avoid the confusing term hemangioendothelioma. If a vascular lesion is benign, it should be classified as a variant of hemangioma. If malignant as angiosarcoma, use a separate category, in which lesions like angioblastoma and angioblastomatosis can be put until their nature has been clarified.

Adult↗

PubMed2003

Discriminant validity and relative precision for classifying patients with nonspecific neck and back pain by anatomic pain patterns.

STUDY DESIGN: Secondary analysis of a previously described cohort of prospective, consecutive patients with acute neck or low back pain referred to outpatient rehabilitation was performed. OBJECTIVE: To estimate discriminant validity and relative precision of two classification procedures (first visit vs multiple visit) in discriminating short-term pain intensity and perceived disability outcomes. SUMMARY OF BACKGROUND DATA: Centralization and noncentralization are pain responses used to classify patients and predict outcomes. Different time frames have been proposed for operationally defining these responses, which are problematic for comparing outcomes across clinical trials. Classifying patients according to pain response observed from initial examination (first visit) and over time (multiple visits) influences prevalence within categories and interpretation of classification usefulness, which merits further investigation. METHODS: Patients with acute onset of nonspecific neck or low back pain referred to two outpatient physical therapy clinics completed body pain diagrams, pain intensity ratings, and disability questionnaires at initial evaluation, during each visit, and at discharge. Therapists collected data enabling patient classification on initial examination and throughout treatment. Differences in pain and disability from intake to discharge from rehabilitation across classification categories were used to assess discriminant validity. Relative precision was estimated by determining ratios of analysis of covariance F values between classification procedures for pain and disability. RESULTS: Both classification procedures discriminated categories for change in pain and disability. The multiple-visit classification procedure was more precise for discriminating outcomes than the first-visit classification procedure. CONCLUSION: Multiple-visit classification of patients into specific pain pattern subgroups is recommended when pain intensity and disability outcomes are of interest.

Adult↗

PubMed1996

Comparative study of the cell tropism of feline immunodeficiency virus isolates of subtypes A, B and D classified on the basis of the env gene V3-V5 sequence.

Feline immunodeficiency virus (FIV) isolates have been classified into subtypes A, B, C and D based on the env gene V3-V5 sequence. The cell tropism of seven new Japanese isolates and a Petaluma (prototype) isolate of FIV, which classified into subtypes A, B and D, for feline lymphoblastoid and feline fibroblastoid cell lines was compared. FeT-1 (CD4+/-, CD8-, AND CD9+2) and Kumi-1 (CD4+2, CD8- and CD9+2) cells were used as the interleukin-2 (IL-2)-dependent feline T-lymphocyte cell lines and FeT-J (CD4+, CD8+/- and CD9+2) and 3201 (CD4+2, CD8+ and CD9-) cells were used as the IL-2- independent feline T-lymphocyte cell lines. The feline fibroblastoid cell lines used were Crandell feline kidney (CrFK) and fewf-4 (both CD4-, CD8- and CD9+2) cells. All FIV isolates replicated in all lymphoblastoid cell lines used. All isolates showed the greatest cytopathogenicity for Kumi-1 cells. All isolates replicated even in the CD9-negative 3201 cells. More isolates caused persistent infection in IL-2-independent cell lines than in IL-2-dependent cell lines. The number of subtype B isolates that established persistent infection was limited, only one of four strains. Only the subtype A isolates replicated in CrFK cells, whereas none of the isolates replicated in fewf-4 cells, which have similar cell surface markers to CrFK cells. The subtype A viruses (CrFK/Petaluma, CrFK/Sendai-1) growing in CrFK cells showed greater cytopathogenicity for lymphoblastoid cell lines than did those (FL-4/Petaluma, Kumi-1/Sendai-1) growing in a lymphoblastoid cell line.

Animals↗

PubMed2001

Statistical mechanics of learning with soft margin classifiers.

We study the typical learning properties of the recently introduced soft margin classifiers (SMCs), learning realizable and unrealizable tasks, with the tools of statistical mechanics. We derive analytically the behavior of the learning curves in the regime of very large training sets. We obtain exponential and power laws for the decay of the generalization error towards the asymptotic value, depending on the task and on general characteristics of the distribution of stabilities of the patterns to be learned. The optimal learning curves of the SMCs, which give the minimal generalization error, are obtained by tuning the coefficient controlling the trade-off between the error and the regularization terms in the cost function. If the task is realizable by the SMC, the optimal performance is better than that of a hard margin support vector machine and is very close to that of a Bayesian classifier.

Algorithms↗

PubMed2006

Integrated state evaluation for the images of crystallization droplets utilizing linear and nonlinear classifiers.

In a usual crystallization process, the researchers evaluate the protein crystallization growth states based on visual impressions and repeatedly assign scores throughout the growth process. Although the development of crystallization robotic systems has generally realised the automation of the setup and storage of crystallization samples, evaluation of crystallization states has not yet been completely automated. The method presented here attempts to categorize individual crystallization droplet images into five classes using multiple classifiers. In particular, linear and nonlinear classifiers are utilized. The algorithm is comprised of pre-processing, feature extraction from images using texture analysis and a categorization process using linear discriminant analysis (LDA) and support vector machine (SVM). The performance of this method has been evaluated by comparing the results obtained using the method with the results obtained by a human expert and the concordance rate was 84.4%.

Algorithms↗