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 289 records · Page 16Linked to original sources
PubMed2005

Learning-induced synchronization and plasticity of a developing neural network.

Learning-induced synchronization of a neural network at various developing stages is studied by computer simulations using a pulse-coupled neural network model in which the neuronal activity is simulated by a one-dimensional map. Two types of Hebbian plasticity rules are investigated and their differences are compared. For both models, our simulations show a logarithmic increase in the synchronous firing frequency of the network with the culturing time of the neural network. This result is consistent with recent experimental observations. To investigate how to control the synchronization behavior of a neural network after learning, we compare the occurrence of synchronization for four networks with different designed patterns under the influence of an external signal. The effect of such a signal on the network activity highly depends on the number of connections between neurons. We discuss the synaptic plasticity and enhancement effects for a random network after learning at various developing stages.

Algorithms↗
PubMed2005

Selection of predictor variables for pneumonia using neural networks and genetic algorithms.

BACKGROUND: Artificial neural networks (ANN) can be used to select sets of predictor variable that incorporate nonlinear interactions between variables. We used a genetic algorithm, with selection based on maximizing network accuracy and minimizing network input-layer cardinality, to evolve parsimonious sets of variables for predicting community-acquired pneumonia among patients with respiratory complaints. METHODS: ANN were trained on data from 1044 patients in a training cohort, and were applied to 116 patients in a testing cohort. Chromosomes with binary genes representing input-layer variables were operated on by crossover recombination, mutation, and probabilistic selection based on a fitness function incorporating both network accuracy and input-layer cardinality. RESULTS: The genetic algorithm evolved best 10-variable sets that discriminated pneumonia in the training cohort (ROC areas, 0.838 for selection based on average cross entropy (ENT); 0.954 for selection based on ROC area (ROC)), and in the testing cohort (ROC areas, 0.847 for ENT selection; 0.963 for ROC selection), with no significant differences between cohorts. Best variable sets based on the genetic algorithm using ROC selection discriminated pneumonia more accurately than variable sets based on stepwise neural networks (ROC areas, 0.954 versus 0.879, p = 0.030), or stepwise logistic regression (ROC areas, 0.954 versus 0.830, p = 0.000). Variable sets of lower cardinalities were also evolved, which also accurately discriminated pneumonia. CONCLUSION: Variable sets derived using a genetic algorithm for neural networks accurately discriminated pneumonia from other respiratory conditions, and did so with greater accuracy than variables derived using stepwise neural networks or logistic regression in some cases.

Algorithms↗
PubMed2005

Automatic determination of synergies by radial basis function artificial neural networks for the control of a neural prosthesis.

This paper describes an automatic method for synthesizing the control for a neural prosthesis (NP) that could augment elbow flexion/extension and forearm pronation/supination in persons with hemiplegia. The basis for the control was a synergistic model of reaching and grasping that uses temporal and spatial synergies between the arm and body segments. The synergies were determined from the movement data measured in nondisabled persons during the performance of functional tasks. The work space was divided into six zones: distance (two attributes) and laterality (three attributes). Radial basis function artificial neural networks (RBF ANN) were used to determine synergies. Sets of RBF ANN characterized with good generalization were selected as control laws for elbow flexion/extension and forearm pronation/supination. The validation was performed for three categories: inter-subject, distance, and laterality generalization. For all of the defined spatial synergies, the correlation was high for inter-subject and distance, yet low for the laterality scenario. This suggests the necessity for implementing different maps for different directions, but the same maps for different distances. The natural movements of the upper arm then drive the lower arm (elbow flexion/extension and forearm pronation/supination) in a way that is very well suited for the administration of functional electrical therapy (FET) in persons with hemiplegia soon after the onset of impairment.

Activities of Daily Living↗
PubMed1999

A micropopulational modelling of a viral epidemic by using a special neural network.

A general forward neural network was adapted for a simulation of viral epidemics. This involves the introduction of a strongly dependence upon history, upon the cumulated values of the corresponding neuron (individual) activations (states of infection) specifying the activation (health) states of the contaminated individuals, represented by the activated neurons and the dynamic parameters of the neural network: the matrix of the synaptic connection and the vector of the activation thresholds (corresponding to the matrix of the viral transfers between the various individuals and to the vector of the minimal individual contamination doses of virus). The recurrence relations and the learning procedures were also adapted to these processes. This methodology was used for the study of the micropopulational spreading of viral epidemics in various epidemiological situations.

Computer Simulation↗
PubMed1995

Quantitative structure-pharmacokinetic relationships (QSPR) of beta blockers derived using neural networks.

This study demonstrates the application of neural networks to predict the pharmacokinetic properties of beta-adrenoreceptor antagonists in humans. A congeneric series of 10 beta-blockers, whose critical pharmacokinetic parameters are well established, was selected for the study. An appropriate neural network system was constructed and tested for its ability to predict the pharmacokinetic parameters from the octanol/water partition coefficient (shake flask method), the pKa, or the fraction bound to plasma proteins. Neural networks successfully trained and the predicted pharmacokinetic values agreed well with the experimental values (average difference = 8%). The neural network-predicted values showed better agreement with the experimental values than those predicted by multiple regression techniques (average difference = 47%). Because the neural networks had a large number of connections, two tests were conducted to determine if the networks were memorizing rather than generalizing. The "leave-one-out" method verified the generalization of the networks by demonstrating that any of the compounds could be deleted from the training set and its value correctly predicted by the new network (average error = 19%). The second test involved the prediction of pharmacokinetic properties of compounds never seen by the network, and reasonable results were obtained for three out of four compounds tested. The results indicate neural networks can be a powerful tool in exploration of quantitative structure-pharmacokinetic relationships.

Acebutolol↗
PubMed1994

Neural networks and psychiatry: candidate applications in clinical decision making.

Neural networks comprise a fundamentally new type of computer system inspired by the functioning of neurons in the brain. Such networks are good at solving problems that involve pattern recognition and categorisation. An important difference between a neural network and a traditional computer system is that in developing an application, a neural network is not programmed; instead, it is trained to solve a particular type of problem. This ability to learn to solve a problem makes neural networks adaptable to solving a wide variety of problems, some of which have proved intractable using a traditional computing approach. Neural networks are particularly suited to tasks involving the categorisation of patterns of information, such as is required in diagnosis and clinical decision making. In the last three years reports of applications involving neural networks have begun to appear in the medical literature, and these are described in this paper. However, a comprehensive search of the literature has shown that there have not as yet been reports of any applications in psychiatry. This paper discusses the nature of clinical decision making, outlines the sorts of problems in psychiatry which neural networks applications might be developed to address, and gives examples of candidate applications in clinical decision making.

Adult↗
PubMed2005

Global exponential stability of delayed competitive neural networks with different time scales.

A competitive neural network model was recently proposed to describe the dynamics of cortical maps, where there are two types of memories: long-term and short-term memories. Such a network is characterized by a system of differential equations with two types of variables, one models the fast neural activity and the other models the slow modification of synaptic strength. In this paper, we introduce a time delay parameter into the neural network model to characterize the signal transmission delays in real neural systems and the finite switch speed in the circuit implementations of neural networks. Then, we analyze the global exponential stability of the delayed competitive neural networks with different time scales. We allow the model has non-differentiable and unbounded functions, and use the nonsmooth analysis techniques to prove the existence and uniqueness of the equilibrium, and derive a new sufficient condition ensuring global exponential stability of the networks.

Animals↗
PubMed1997

Ventilation mode recognition using artificial neural networks.

This study investigated the capabilities of artificial neural networks to identify spontaneous and pressure support ventilation modes from gas flow and airway pressure signals. After receiving written informed consent, flow and pressure waveforms were recorded from 13 patients undergoing general anesthesia. During analysis, the inspiratory phase of each breath was extracted and normalized in amplitude and wavelength. Neural networks were configured to input flow, pressure, or both waveforms and to output the ventilatory mode. Neural network training was accomplished with data from 500 breaths obtained from 7 patients. Neural network performance was tested with 433 breaths from the remaining 6 patients. Networks using flow, pressure, and both waveforms recognized correctly 78% (337), 97% (423), and 100% (433) of the test waveforms, respectively. Results indicate that neural networks can be used effectively for breathing pattern recognition and encourage the application of neural networks in other types of respiratory pattern recognition problems.

Aged↗
PubMed1994

Artificial neural networks within medical decision support systems.

Artificial neural networks offer a way to actively assimilate both past and present knowledge, to extract information, to map correlations and to produce inferences from available data; all tasks which have relevance to the clinical laboratory. In this paper, we describe one useful artificial neural network technique, backpropagation, and describe some of the practical considerations which need to be taken account of when using such methods. Examples are presented of the application of artificial neural networks in medicine and, particularly, in clinical chemistry. The paper goes on to describe the use of these methods within medical decision support. We conclude that artificial neural networks are useful multivariate techniques which are well able to play an important role in a decision support system. Further, that their properties as function approximators could be utilised in other areas of clinical chemistry. We conclude by pointing out that the pattern recognition ability of artificial neural networks holds out the promise of extracting useful information from currently available data which is at present seen as being of little diagnostic utility.

Chemistry, Clinical↗
PubMed1994

On the classification capability of a dynamic threshold neural network.

This paper proposes a new type of neural network called the Dynamic Threshold Neural Network (DTNN) which is theoretically and experimentally superior to a conventional sigmoidal multilayer neural network in classification capability. Given a training set containing 4k + 1 patterns in Rn, to successfully learn this training set, the upper bound on the number of free parameters for a DTNN is (k + 1)(n + 2) + 2(k + 1), while the upper bound for a sigmoidal network is 2k(n + 1) + (2k + 1). We also derive a learning algorithm for the DTNN in a similar way to the derivation of the backprop learning algorithm. In simulations on learning the Two-Spirals problems, our DTNN with 30 neurons in one hidden layer takes only 3200 epochs on average to successfully learn the whole training set, while the single-hidden-layer feedforward sigmoidal neural networks have never been reported to successfully learn the given training set even though more hidden neurons are used.

Algorithms↗
PubMed2000

Automatic detection and quantification of ground-glass opacities on high-resolution CT using multiple neural networks: comparison with a density mask.

OBJECTIVE: We compared multiple neural networks with a density mask for the automatic detection and quantification of ground-glass opacities on high-resolution CT under clinical conditions. SUBJECTS AND METHODS: Eighty-four patients (54 men and 30 women; age range, 18-82 years; mean age, 49 years) with a total of 99 consecutive high-resolution CT scans were enrolled in the study. The neural network was designed to detect ground-glass opacities with high sensitivity and to omit air-tissue interfaces to increase specificity. The results of the neural network were compared with those of a density mask (thresholds, -750/-300 H), with a radiologist serving as the gold standard. RESULTS: The neural network classified 6% of the total lung area as ground-glass opacities. The density mask failed to detect 1.3%, and this percentage represented the increase in sensitivity that was achieved by the neural network. The density mask identified another 17.3% of the total lung area to be ground-glass opacities that were not detected by the neural network. This area represented the increase in specificity achieved by the neural network. Related to the extent of the ground-glass opacities as classified by the radiologist, the neural network (density mask) reached a sensitivity of 99% (89%), specificity of 83% (55%), positive predictive value of 78% (18%), negative predictive value of 99% (98%), and accuracy of 89% (58%). CONCLUSION: Automatic segmentation and quantification of ground-glass opacities on high-resolution CT by a neural network are sufficiently accurate to be implemented for the preinterpretation of images in a clinical environment; it is superior to a double-threshold density mask.

Absorptiometry, Photon↗
PubMed1995

Prediction of the early prognosis of the hepatectomized patient with hepatocellular carcinoma with a neural network.

The early prognosis of the hepatectomized patients with hepatocellular carcinoma was determined preoperatively with a perceptron-type neural network. The neural network was trained with the preoperative data of 54 example cases with the early prognosis, successful or died of hepatic dysfunction, as teaching signals. After learning these examples, the neural network came to give a precise prediction to the example data except for one case. With the learned neural network, the outcomes of the hepatectomy of 11 patients (10 successful; 1 died) were predicted prospectively with 100% precision. The usefulness of the neural network for the prediction was determined.

Adult↗
PubMedDate not supplied

Prediction of Henry's law constants by a quantitative structure property relationship and neural networks.

Multiple linear regression analysis and neural networks were employed to develop predictive models for Henry's law constants (HLCs) for organic compounds of environmental concern in pure water at 25 degrees C, using a set of quantitative structure property relationship (QSPR)-based descriptors to encode various molecular structural features. Two estimation models were developed from a set of 303 compounds using 10 and 12 descriptors, one of these models using two descriptors to account for hydrogen-bonding characteristics explicitly; these were validated subsequently on an external set of 54 compounds. For each model, a linear regression and neural network version was prepared. The standard errors of the linear regression models for the training data set were 0.262 and 0.488 log(H(cc)) units, while those of the neural network analogues were lower at 0.202 and 0.224, respectively; the linear regression models explained 98.3% and 94.3% of the variance in the development data, respectively, the neural network models giving similar quality results of 99% and 98.3%, respectively. The various descriptors used describe connectivity, charge distribution, charged surface area, hydrogen-bonding characteristics, and group influences on HLC values.

Journal Article↗
PubMed1994

Optimization dynamics for partitioned neural networks.

Given a relaxation-based neural network and a desired partition of the neurons in the network into modules with relatively slow communication between modules, we investigate relaxation dynamics for the resulting partitioned neural network. In particular, we show how the slow inter-module communication channels can be modeled by means of certain transformations of the original objective function which introduce new state variables for the inter-module communication links. We report on a parallel implementation of the resulting relaxation dynamics, for a two-dimensional image segmentation network, using a network of workstations. Experiments demonstrate a functional and efficient parallelization of this neural network algorithm. We also discuss implications for analog hardware implementations of relaxation networks.

Kinetics↗
PubMed2000

Scandinavian test of artificial neural network for classification of myocardial perfusion images.

Artificial neural networks are systems of elementary computing units capable of learning from examples. They have been applied to automated interpretation of myocardial perfusion images and have been shown to perform even better than experienced physicians. It has been shown that physicians interpreting myocardial perfusion images benefit from the advice of such networks. These networks have been developed and validated in the same hospital. However, widespread use of neural networks will only take place if the networks can maintain a high accuracy in other hospitals, i.e. hospitals using different gamma cameras, different acquisition techniques, different study protocols, etc. The purpose of this study was to develop a neural network in one hospital and test it in another. An artificial neural network was trained to detect coronary artery disease using myocardial perfusion scintigrams from 135 patients at a Swedish hospital. Thereafter, this network was tested using scintigrams from 68 patients at a Danish hospital and compared to six criteria based on expert physician analysis and quantitative analysis by the CEqual program. The sensitivity of the network was significantly higher than that of one of the physician criteria (0. 92 versus 0.71) and two of the CEqual-based criteria (0.94 versus 0. 63 and 0.96 versus 0.65) compared at equal specificities. It was concluded that an artificial neural network can maintain high accuracy in a hospital other than the one where it was developed.

Coronary Disease↗
PubMedDate not supplied

Supervised feature ranking using a genetic algorithm optimized artificial neural network.

A genetic algorithm optimized artificial neural network GNW has been designed to rank features for two diversified multivariate data sets. The dimensions of these data sets are 85x24 and 62x25 for 24 or 25 molecular descriptors being computed for 85 matrix metalloproteinase-1 inhibitors or 62 hepatitis C virus NS3 protease inhibitors, respectively. Each molecular descriptor computed is treated as a feature and input into an input layer node of the artificial neural network. To optimize the artificial neural network by the genetic algorithm, each interconnected weight between input and hidden or between hidden and output layer nodes is binary encoded as a 16 bits string in a chromosome, and the chromosome is evolved by crossover and mutation operations. Each input layer node and its associated weights of the trained GNW are systematically omitted once (the self-depleted weights), and the corresponding weight adjustments due to the omission are computed to keep the overall network behavior unchanged. The primary feature ranking index defined as the sum of self-depleted weights and the corresponding weight adjustments computed is found capable of separating good from bad features for some artificial data sets of known feature rankings tested. The final feature indexes used to rank the data sets are computed as a sum of the weighted frequency of each feature being ranked in a particular rank for each data set being partitioned into numerous clusters. The two data sets are also clustered by a standard K-means method and trained by a support vector machine (SVM) for feature ranking using the computed F-scores as feature ranking index. It is found that GNW outperforms the SVM method on three artificial as well as the matrix metalloproteinase-1 inhibitor data sets studied. A clear-cut separation of good from bad features is offered by the GNW but not by the SVM method for a feature pool of known feature ranking.

Algorithms↗
PubMed2002

On the complexity of computing and learning with multiplicative neural networks.

In a great variety of neuron models, neural inputs are combined using the summing operation. We introduce the concept of multiplicative neural networks that contain units that multiply their inputs instead of summing them and thus allow inputs to interact nonlinearly. The class of multiplicative neural networks comprises such widely known and well-studied network types as higher-order networks and product unit networks. We investigate the complexity of computing and learning for multiplicative neural networks. In particular, we derive upper and lower bounds on the Vapnik-Chervonenkis (VC) dimension and the pseudo-dimension for various types of networks with multiplicative units. As the most general case, we consider feedforward networks consisting of product and sigmoidal units, showing that their pseudo-dimension is bounded from above by a polynomial with the same order of magnitude as the currently best-known bound for purely sigmoidal networks. Moreover, we show that this bound holds even when the unit type, product or sigmoidal, may be learned. Crucial for these results are calculations of solution set components bounds for new network classes. As to lower bounds, we construct product unit networks of fixed depth with super-linear VC dimension. For sigmoidal networks of higher order, we establish polynomial bounds that, in contrast to previous results, do not involve any restriction of the network order. We further consider various classes of higher-order units, also known as sigma-pi units, that are characterized by connectivity constraints. In terms of these, we derive some asymptotically tight bounds. Multiplication plays an important role in both neural modeling of biological behavior and computing and learning with artificial neural networks. We briefly survey research in biology and in applications where multiplication is considered an essential computational element. The results we present here provide new tools for assessing the impact of multiplication on the computational power and the learning capabilities of neural networks.

Artificial Intelligence↗
PubMed1995

Identification of ribosome binding sites in Escherichia coli using neural network models.

This study investigated the use of neural networks in the identification of Escherichia coli ribosome binding sites. The recognition of these sites based on primary sequence data is difficult due to the multiple determinants that define them. Additionally, secondary structure plays a significant role in the determination of the site and this information is difficult to include in the models. Efforts to solve this problem have so far yielded poor results. A new compilation of E. coli ribosome binding sites was generated for this study. Feedforward backpropagation networks were applied to their identification. Perceptrons were also applied, since they have been the previous best method since 1982. Evaluation of performance for all the neural networks and perceptrons was determined by ROC analysis. The neural network provided significant improvement in the recognition of these sites when compared with the previous best method, finding less than half the number of false positives when both models were adjusted to find an equal number of actual sites. The best neural network used an input window of 101 nucleotides and a single hidden layer of 9 units. Both the neural network and the perceptron trained on the new compilation performed better than the original perceptron published by Stormo et al. in 1982.

Base Sequence↗