Search PubMed⌕ Search

SEARCH · Search PubMed

Results for “Neural network”

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 1,549 records · Page 86Linked to original sources
PubMed1999

Analysis of the symptoms of depression--a neural network approach

The purpose of this study is to determine the individual contribution, or importance number, of the symptoms to an analysis of depression, utilizing a neural network model. In addition, the presence of hopelessness and somatic complaints was examined, to determine their relevance to depression. Using Wave 1 data from Duke University's contribution in the Epidemiological Catchment Area (ECA) study, we created a mathematical model, a neural network, to map the relationship of nine symptoms of major depression, hopelessness and somatic complaints to the presence or absence of the formal diagnosis of depression, and performed a contribution analysis. The contribution analysis using the neural network revealed that the symptoms with the greatest impact on the occurrence of depression, or with the largest importance number for depression, were sadness, loss of interest, tiredness and sleeping trouble, in that order. The most frequently reported symptoms, though, were sadness, sleeping trouble, suicidal ideation, tiredness and poor concentration, in that order. Hopelessness and somatic symptoms were the lowest in their contribution to the diagnosis of depression. The study thus provides the hierarchy of the symptoms of depression and supports the DSM classification of major depression.

Journal Article↗
PubMedDate not supplied

Is it possible to predict one-year survival in centenarians? A neural network study.

BACKGROUND: Human life expectancy is constantly increasing: the challenge for modern geriatric medicine is to identify the means to reach successfully extreme longevity. OBJECTIVE: To determine which are the survival determinants in centenarians using a neural network. METHODS: Sample of 110 centenarians living in Rome, mean age 101.6 years (SD=1.8) with a sex ratio males:females of 1:3. We administered an extensive health interview (lasting 1-2 h) to each subject. The questionnaire, carried out according to the Geriatric Multidimensional Assessment, is made up of 100 items including a comprehensive health and psychosocial assessment aimed at various topics of general health and well-being and some scales used in geriatric practice. We applied several three-layered feed-forward neural networks by mixing in different ways the most important of the 100 items. RESULTS: The most predicting powered net is the one constructed with 23 variables regarding comorbidity, cardiovascular risk factors, cognitive status, mood, functional status and social interactions, which therefore are strictly related to survival in centenarians. CONCLUSION: Survival in longevity is a complex biological phenomenon, which is an ideal field for using the neural network as a statistic method. The net shows us that the maintenance of social relationships even in presence of disability is of major importance for survival in the oldest old.

Aged, 80 and over↗
PubMedDate not supplied

Artificial neural networks can distinguish novice and expert strategies during complex problem solving.

OBJECTIVE: To determine whether expert problem-solving strategies can be identified within a large number of student performances of complex medical diagnostic simulations. METHODS: Self-organizing artificial neural networks were trained to categorize the performances of infectious disease subspecialists on six computer-based clinical diagnostic simulation that used the sequence of diagnostic tests requested as the input data. Six hundred seventy-six student solutions to these problems were presented to these trained neural networks to determine which, if any, of the student solutions represented those of the experts. RESULTS: For each simulation, the expert performances clustered around one dominant output neurode, indicating that there were common problem-specific features associated with the experts' problem-solving performances. When the performances of students who also made correct problem diagnoses were tested on these expert-trained neural networks, 17% were classified as representing expert strategies, indicating that expert performance was a somewhat rare and inconsistent occurrence among the students. CONCLUSIONS: The ability to identify a small number of expert-like strategies within a large body of student performances may provide an opportunity to study the dynamics of complex learning at both individual and population levels as well as the emergence of medical diagnostic expertise.

Diagnostic Tests, Routine↗
PubMed1997

Neural networks as 'software sensors' in enzyme production.

Industrial applications of enzyme technology are rapidly increasing. On-line control of enzyme production processes, however, is difficult, owing to the uncertainties typical of biological reactions and to the lack of suitable sensors. We demonstrate that well-trained feedforward backpropagation neural networks with one hidden layer can be employed to overcome such problems with no need for a priori knowledge of the relationships of the process variables involved. Neural network programs were written in Microsoft Visual C+2 for Windows and implemented in a personal computer. The goodness of fit of the trained neural network to the reference data was determined by the coefficient of determination R2. On-line state estimation and multi-step ahead prediction of enzyme activity and biomass concentration, both in a yeast lipase and fungal glucoamylase production could be satisfactorily carried out. Results showed an excellent fit for estimated lipase activity (R2 = 0.988) and biomass concentration (R2 = 0.989). In glucoamylase production, both enzyme activity and biomass concentration could also be reliably predicted for 2 time intervals (10 h) ahead with only on-line measurable parameter values as the input data.

Algorithms↗
PubMed1992

Sleep classification in infants based on artificial neural networks.

The study reports on the possibility of classifying sleep stages in infants using an artificial neural network. The polygraphic data from 4 babies aged 6 weeks, 6 months and 1 year recorded over 8 hours were available for classification. From each baby 22 signals were recorded, digitized and stored on an optical disc. Subsets of these signals and additional calculated parameters were used to obtain data vectors, each of which represents an interval of 30 sec. For classification, two types of neural networks were used, a Multilayer Perceptron and a Learning Vector Quantizer. The teaching input for both networks was provided by a human expert. For the 6 sleep classes in babies aged 6 months, a 65% to 80% rate of correct classification (4 babies) was obtained for the testing data not previously seen.

Cerebral Cortex↗
PubMed2002

[Chemical QSAR recognition by using fuzzy min-max neural-network].

By using the fuzzy min-max neural network, the quantitative structure-activity relationship (QSAR) of mutagenicity is studied. With the established QSAR model, the mutagenicity is predicted and the results showed that QASR is superior to linear-regression model. Further discussion on the models and the results is presented in this paper.

Algorithms↗
PubMed2002

Prediction of emergency department visits for respiratory symptoms using an artificial neural network.

STUDY OBJECTIVES: Accurate prediction of the effect of atmospheric changes, including pollutants, on emergency department (ED) visits for respiratory symptoms would be useful, but has proven difficult. The main difficulty is the limitation of the classical linear models and logistic regression with multiple variables to handle the multifactorial effect. DESIGN AND SETTING: To predict ED visits, we have created a computer-based model called an artificial neural network (ANN) using a back-propagation training algorithm and genetic algorithm optimization. This ANN was fed meteorologic and air pollution input variables and trained to predict the number of patients admitted to the ED with respiratory symptoms of asthma, COPD, and acute and chronic bronchitis on the corresponding day. One thousand twenty data sets were extracted from an ED admittance database at the Barzilai Medical Center (Ashkelon, Israel), and randomized to a network training set (n = 816) and a test set (n = 204). RESULTS: The neural network performed best when the predictor variables used were temperature, relative humidity, barometric pressure, SO(2), and oxidation products of nitric oxide, and the data presented as peak value 24 h prior to ED admission and the average during the 7 days before the ED visit. The neural network was able to predict the test set with an average error of 12%. CONCLUSION: Based on meteorologic and pollution data, the use of an ANN can assist in the prediction of ED visits related to respiratory conditions.

Algorithms↗
PubMed2002

Exponential stability of Cohen-Grossberg neural networks.

Exponential stabilities of the Cohen-Grossberg neural network with and without delays are analyzed. By Liapunov functions/functionals, sufficient conditions are obtained for general exponential stability, while by using a comparison result from the theory of monotone dynamical systems, componentwise exponential stability is also discussed. All results are established without assuming any symmetry of the connection matrix, and the differentiability and monotonicity of the activation functions.

Neural Networks, Computer↗
PubMed1997

Classification of factor deficiencies from coagulation assays using neural networks.

Activated partial thromboplastin time (APTT) and prothrombin time (PT) assays are widely used to screen for coagulation disorders and to monitor administration of therapeutic drugs. The analysis of data from coagulation assays has traditionally concentrated on determination of clot times (for APTT and PT) and magnitude of signal change during coagulation (e.g. for PT-based fibrinogen quantitation). The purpose of this study was to determine if the diagnostic power of these assays could be increased by using neural networks to interpret multiple parameters from these assays. Error back-propagation neural networks were trained using multiple variables derived from APTT and PT optical data for 200 normal and abnormal patient specimens. These networks were used to: (1) classify samples as either deficient or non-deficient with respect to individual blood components; and (2) estimate the approximate concentration of specific coagulation factors. Results indicated that these networks could be successfully trained to identify specific factor deficiencies at less than 30% normal levels with good specificity and variable sensitivity, but that they estimated actual concentrations poorly in most cases. These results support possible applications for neural networks identifying specific coagulation abnormalities from non-specific APTT and PT assays using expanded data parameter sets.

Blood Coagulation Disorders↗
PubMed2003

[Prediction of human skin permeability of drugs in vivo with artificial neural network].

OBJECTIVE: To predict in vivo human skin permeability of drugs. METHODS: Appropriate BP(Back-propagation) neural network to predict human skin permeation ratios of drugs (R, absorbed/unabsorbed) was established. The octanol water partition coefficients (logP), molecular volumes (V), hydrogen bond acidities (sigma alpha 2(H)) and hydrogen bond basidities (sigma beta 2(H)) were selected as the neural units of input layer, and logR values were selected as the neural units of output layer. RESULTS: The calculated logR values of 17 drugs were in good accordance with their observed values. CONCLUSION: BP neural network can be used to predict in vivo human skin permeability of drugs.

Humans↗
PubMed2005

Passivity-based neural network adaptive output feedback control for nonlinear nonnegative dynamical systems.

The potential clinical applications of adaptive neural network control for pharmacology in general, and anesthesia and critical care unit medicine in particular, are clearly apparent. Specifically, monitoring and controlling the depth of anesthesia in surgery is of particular importance. Nonnegative and compartmental models provide a broad framework for biological and physiological systems, including clinical pharmacology, and are well suited for developing models for closed-loop control of drug administration. In this paper, we develop a neural adaptive output feedback control framework for adaptive set-point regulation of nonlinear uncertain nonnegative and compartmental systems. The proposed framework is Lyapunov-based and guarantees ultimate boundedness of the error signals corresponding to the physical system states and the neural network weighting gains. The approach is applicable to nonlinear nonnegative systems with unmodeled dynamics of unknown dimension and guarantees that the physical system states remain in the nonnegatiye orthant of the state-space for nonnegative initial conditions. Finally, a numerical example involving the infusion of the anesthetic drug midazolam for maintaining a desired constant level of depth of anesthesia for noncardiac surgery is provided to demonstrate the efficacy of the proposed approach.

Adaptation, Physiological↗
PubMed2005

Biological engineering applications of feedforward neural networks designed and parameterized by genetic algorithms.

Two neural network (NN) applications in the field of biological engineering are developed, designed and parameterized by an evolutionary method based on the evolutionary process of genetic algorithms. The developed systems are a fault detection NN model and a predictive modeling NN system. An indirect or 'weak specification' representation was used for the encoding of NN topologies and training parameters into genes of the genetic algorithm (GA). Some a priori knowledge of the demands in network topology for specific application cases is required by this approach, so that the infinite search space of the problem is limited to some reasonable degree. Both one-hidden-layer and two-hidden-layer network architectures were explored by the GA. Except for the network architecture, each gene of the GA also encoded the type of activation functions in both hidden and output nodes of the NN and the type of minimization algorithm that was used by the backpropagation algorithm for the training of the NN. Both models achieved satisfactory performance, while the GA system proved to be a powerful tool that can successfully replace the problematic trial-and-error approach that is usually used for these tasks.

Algorithms↗
PubMed1992

Relationships between secondary structure fractions for globular proteins. Neural network analyses of crystallographic data sets.

The relationship between the fractions of protein secondary structural components as determined from X-ray crystallographic data by the procedures of Kabsch and Sander (KS) and of Levitt and Greer (LG) is analyzed by neural network analysis of these two tabulations of literature data. A linear relationship between the KS and LG reductions of X-ray data to secondary structure descriptors is demonstrated by a regression analysis of the relationships between these sets of structural parameters. Back-propagation neural network analysis was then used to derive equations for determination of the most probable fractions of beta-sheet, bend, turn, and "other" conformations given the fraction of alpha-helix in a globular protein. The deviation of the X-ray values for beta-sheet from that determined with these equations was shown to have a variance that exponentially decreased with increasing fraction of alpha-helix. A second neural network analysis showed that knowledge of both the alpha-helical and beta-sheet fractions in a protein significantly reduces the uncertainty in prediction of the other components of the secondary structure. These analyses provide insight into the nature of the data sets derived from crystal structures. Since these complications of crystal structure data are commonly used as reference information for quantitative evaluation of spectra (for example, FTIR, Raman, and electronic or vibrational circular dichroism) in terms of secondary structure, such internal correlations in the reference sets may have significant effects on the stability of spectroscopic analyses derived from them.

Mathematics↗
PubMed1998

Classification and quantitation of 1H NMR spectra of alditols binary mixtures using artificial neural networks.

A pattern recognition method based on artificial neural networks (ANNs) to analyze and quantify the components of six alditol binary mixtures is presented. This method is suitable to classify the spectra of the 15 mixtures obtained from the six alditols and to produce quantitative estimates of the component concentrations. The system is user-friendly and is helpful in solving the problem of greatly overlapping signals, often encountered in NMR spectroscopy of carbohydrates. A "classification" ANN uses 200 intensity values of the 1H NMR spectrum in the range 3.5-4 ppm. When the correct mixture is identified, the quantification is solved by assigning a specific ANN to each mixture. These ANNs use the same 200 values of the spectrum and output the values of the two concentrations. The error in the ANN responses is studied, and a method is developed to estimate the accuracy in determining the concentrations. The networks' abilities to recognize previously unseen mixtures are tested. When the classification ANN (trained on the 15 binary mixtures) is exposed to complex (i.e., more than binary) mixtures of the six known alditols, it successfully identifies the components if their minimum concentration is 10%. Given the precision of the results and the small number of errors reported, we believe that the method can be used in all fields in which the recognition and quantification of components are necessary.

Galactitol↗
PubMed2001

Fingerprints classification using artificial neural networks: a combined structural and statistical approach.

This paper describes a fingerprint classification algorithm using Artificial Neural Networks (ANN). Fingerprints are classified into six categories: arches, tented arches, left loops, right loops, whorls and twin loops. The algorithm extracts a string of symbols using the block directional image of a fingerprint, which represents the set of structural features for this image. The moment representing the statistical feature of the pattern is computed for this string and its Euclidean Distance Measures (EDM) are computed by using this moment. Our discrimination system uses a multilayer artificial neural network composed of six subnetworks one for each class. The classifier was tested on 1,500 images of good quality in the Egyptian Fingerprints database; images with poor quality were rejected. In the six-class problem the network achieved 95% classification accuracy. In the five-class problem when we place whorls and twin loops together in the same category the classification accuracy was around 99%. In the four-class problem when we place arches and tented arches in the same class the classification accuracy was 99%.

Algorithms↗
PubMed1997

Use of neural networks in telemedical monitoring of divers.

An artificial neural network (ANN) has been developed to predict and classify the risk of medical disorders for certain dive profiles. The telemedical data applied to the ANN represent different physiological and physical parameters that are transmitted acoustically from a diver to a remote receiver. The telemetered data include dive time, maximum depth and decompression stop times and depths. Preliminary tests demonstrated the successful performance of the ANN where classical mathematical and statistical methods had failed due to the complex nature and variability of the parameters involved.

Diving↗
PubMed1996

Classification of quantitative EEG data by an artificial neural network: a preliminary study.

Previous studies from different laboratories have suggested that qEEG could be useful for distinguishing dementia from normality. Our aims were: (1) to study the ability of qEEG to distinguish dementia among different pathological conditions in ambulatory settings; (2) to compare the ability of classical statistical analysis and of neural networks in classifying qEEG data. We were able to obtain a multiple discriminant function using a training set of patients, which classified correctly more than 91% of the qEEGs from an independent group of patients, with less than 5% of false positives. Kohonen's neural network was trained with the same set of patients. This unsupervised learning artificial neural network performed the classification of the independent sample with an accuracy comparable to that of the multiple discriminant function. Our results suggest that the use of unsupervised learning algorithms could be an interesting alternative in the classification of data obtained from psychiatric patients where definition of their clinical profile is not always a simple task.

Adult↗
PubMed1999

Analysis of complex traits using neural networks.

A recently developed approach that employs artificial neural networks (ANNs) was applied to the simulated data set to identify sets of marker loci involved in disease etiology. In this implementation, ANNs are trained to predict the disease state (output) from the given genetic marker data (input). A contribution value (CV) for each locus is calculated from the weights that represent the strength of the connections for the trained ANN; a higher CV indicates a higher probability of linkage. The highest CV values were chosen as the most likely candidate regions involved in the disease.

Computer Simulation↗