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 523 records · Page 29Linked to original sources
PubMed2004

An annealed chaotic maximum neural network for bipartite subgraph problem.

In this paper, based on maximum neural network, we propose a new parallel algorithm that can help the maximum neural network escape from local minima by including a transient chaotic neurodynamics for bipartite subgraph problem. The goal of the bipartite subgraph problem, which is an NP- complete problem, is to remove the minimum number of edges in a given graph such that the remaining graph is a bipartite graph. Lee et al. presented a parallel algorithm using the maximum neural model (winner-take-all neuron model) for this NP- complete problem. The maximum neural model always guarantees a valid solution and greatly reduces the search space without a burden on the parameter-tuning. However, the model has a tendency to converge to a local minimum easily because it is based on the steepest descent method. By adding a negative self-feedback to the maximum neural network, we proposed a new parallel algorithm that introduces richer and more flexible chaotic dynamics and can prevent the network from getting stuck at local minima. After the chaotic dynamics vanishes, the proposed algorithm is then fundamentally reined by the gradient descent dynamics and usually converges to a stable equilibrium point. The proposed algorithm has the advantages of both the maximum neural network and the chaotic neurodynamics. A large number of instances have been simulated to verify the proposed algorithm. The simulation results show that our algorithm finds the optimum or near-optimum solution for the bipartite subgraph problem superior to that of the best existing parallel algorithms.

Algorithms↗
PubMed1996

Circular nodes in neural networks.

In the usual construction of a neural network, the individual nodes store and transmit real numbers that lie in an interval on the real line; the values are often envisioned as amplitudes. In this article we present a design for a circular node, which is capable of storing and transmitting angular information. We develop the forward and backward propagation formulas for a network containing circular nodes. We show how the use of circular nodes may facilitate the characterization and parameterization of periodic phenomena in general. We describe applications to constructing circular self-maps, periodic compression, and one-dimensional manifold decomposition. We show that a circular node may be used to construct a homeomorphism between a trefoil knot in R3 and a unit circle. We give an application with a network that encodes the dynamic system on the limit cycle of the Kuramoto-Sivashinsky equation. This is achieved by incorporating a circular node in the bottleneck layer of a three-hidden-layer bottleneck network architecture. Exploiting circular nodes systematically offers a neural network alternative to Fourier series decomposition in approximating periodic or almost periodic functions.

Neural Networks, Computer↗
PubMed1997

Comparison of intraocular lens computations using a neural network versus the Holladay formula.

PURPOSE: To compare the accuracy of intraocular lens (IOL) calculations using Holladay personalized calculations and a new method of trained neural networks. SETTING: A private ophthalmic practice. METHODS: We developed and trained a neural network to predict IOL powers using a personalized Holladay program and clinical data from 200 consecutive cases of one surgeon's results with one IOL. Clinical data included preoperative axial length, both keratometry values, anterior chamber depth, and human lens thickness. The neural network was trained to produce the actual postoperative refractive error, and the Holladay surgeon factor was continuously refined using the same results. After the network was successfully trained against the clinical data, it was used to compute IOL power in a double-masked study. Ninety-five patients were randomized between the Holladay personalized calculation and the neural network computation. There were no significant differences in age or preoperative refractive errors between the two groups. Manifest refractions were obtained during the masked period at least 6 weeks after surgery. RESULTS: Mean postoperative error from predicted refraction was +0.271 diopters (D) for the neural network group and -0.217 D for the Holladay personal group. Mean absolute error from predicted refraction was +0.63 D for the neural network group and +0.93 D for the Holladay personal group. There was a significant difference in postoperative refractive errors and mean absolute error between the two groups (P < .022; nonparametric Mann-Whitney test). An error of less than +/- 0.75 D was obtained by 72.5% of the neural network group and 50.0% of the Holladay group. CONCLUSIONS: The neural network prediction formula can improve IOL implantation calculations by tightening the variance of errors.

Cataract Extraction↗
PubMed1998

How embedded memory in recurrent neural network architectures helps learning long-term temporal dependencies.

Learning long-term temporal dependencies with recurrent neural networks can be a difficult problem. It has recently been shown that a class of recurrent neural networks called NARX networks perform much better than conventional recurrent neural networks for learning certain simple long-term dependency problems. The intuitive explanation for this behavior is that the output memories of a NARX network can be manifested as jump-ahead connections in the time-unfolded network. These jump-ahead connections can propagate gradient information more efficiently, thus reducing the sensitivity of the network to long-term dependencies. This work gives empirical justification to our hypothesis that similar improvements in learning long-term dependencies can be achieved with other classes of recurrent neural network axchitectures simply by increasing the order of the embedded memory. In particular we explore the impact of learning simple long-term dependency problems on three classes of recurrent neural network architectures: globally recurrent networks, locally recurrent networks, and NARX (output feedback) networks.Comparing the performance of these architectures with different orders of embedded memory on two simple long-term dependencies problems shows that all of these classes of network architectures demonstrate significant improvement on learning long-term dependencies when the orders of embedded memory are increased. These results can be important to a user comfortable with a specific recurrent neural network architecture because simply increasing the embedding memory order of that architecture will make it more robust to the problem of long-term dependency learning.

Journal Article↗
PubMed2000

Evaluation of neural network models with generalized sensitivity analysis

A sensitivity analysis method for discovering characteristic features of the input data using neural network classification models has been devised. The sensitivity is the gradient of the neural network model response function, and because neural network models are nonlinear, the gradient depends on the point where it is evaluated. Two criteria are used for measuring the sensitivity. The first criterion calculates the sensitivity or gradient of the neural network output with respect to the average of the objects that comprise each class. The second criterion measures the average sensitivity of the class objects. The sensitivity analysis was applied to temperature-constrained cascade correlation network models and evaluated with sets of synthetic data and experimental mobility spectra. The neural network models were built using temperature-constrained cascade correlation networks (TCCCNs). A weight constraint was devised for the output units of the network models. This method implements weight decay with conjugate gradient training and yields more sensitive neural network models. Temperature-constrained hidden units furnish more sensitive network models than networks without constraints. By comparing the sensitivities of the class mean input and the mean sensitivity for all the inputs of a class, the individual input variables may be assessed for linearity. If these two sensitivities for an input variable differ by a constant factor, then that variable is modeled by a simple linear relationship. If the two sensitivities vary by a nonconstant scale factor, then the variable is modeled by higher order functions in the network. The sensitivity method was used to diagnose errors in the training data, and the test for linearity indicated a TCCCN architecture that had better predictability.

Journal Article↗
PubMed1994

Application of artificial neural networks for prokaryotic transcription terminator prediction.

Artificial neural networks (ANN) to predict terminator sequences, based on a feed-forward architecture and trained using the error back propagation technique, have been developed. The network uses two different methods for coding nucleotide sequences. In one the nucleotide bases are coded in binary while the other uses the electron-ion interaction potential values (EIIP) of the nucleotide bases. The latter strategy is new, property based and substantially reduces the network size. The prediction capacity of the artificial neural network using both coding strategies is more than 95%.

DNA↗
PubMed2002

Artificial neural network--mechanism and application in pathology.

Artificial neural network (ANN) is the unique system, which simulates human brain in many respects and there is high hope that it may achieve higher mental functions. In this brief review we have discussed the basic mechanism of artificial neural network along with its application in the field of pathology.

Artificial Intelligence↗
PubMed2004

Prediction of postoperative morbidity after lung resection using an artificial neural network ensemble.

OBJECTIVE: To propose an ensemble model of artificial neural networks (ANNs) to predict cardio-respiratory morbidity after pulmonary resection for non-small cell lung cancer (NSCLC). METHODS: Prospective clinical study was based on 489 NSCLC operated cases. An artificial neural network ensemble was developed using a training set of 348 patients undergoing lung resection between 1994 and 1999. Predictive variables used were: sex of the patient, age, body mass index, ischemic heart disease, cardiac arrhythmia, diabetes mellitus, induction chemotherapy, extent of resection, chest wall resection, perioperative blood transfusion, tumour staging, forced expiratory volume in 1s percent (FEV(1)%), and predicted postoperative FEV(1)% (ppoFEV(1)%). The analysed outcome was the occurrence of postoperative cardio-respiratory complications prospectively recorded and codified. The artificial neural network ensemble consists of 100 backpropagation networks combined via a simple averaging method. The probabilities of complication calculated by ensemble model were obtained to the actual occurrence of complications in 141 cases operated on between January 2000 and December 2001 and a receiver operating characteristic (ROC) curve for this method was constructed. RESULTS: The prevalence of cardio-respiratory morbidity was 0.25 in the training and 0.30 in the validation series. The accuracy for morbidity prediction (area under the ROC curve) was 0.98 by the ensemble model. CONCLUSION: In this series an artificial neural network ensemble offered a high performance to predict postoperative cardio-respiratory morbidity.

Aged↗
PubMedDate not supplied

Automated diagnosis of heart disease in patients with heart murmurs: application of a neural network technique.

This study was conducted to test a three-layered artificial neural network analysis of phonocardiogram recordings to diagnose, automatically and objectively, the condition of the heart in patients with heart murmurs. The data were recorded simultaneously in each of 49 patients with a heart murmur through eight microphones attached to the skin surface with adhesive tape, and were analysed by computer. The diagnosis was automated using a three-layered neural network technique. The neural network generated correct answers in over 70% of cases. Furthermore, about 80% of cases of two concurrent diseases were identified correctly. However, ventricular septal defects were incorrectly classified as aortic stenosis or aortic regurgitation, and patent ductus arteriosus was not diagnosed correctly. Accurate diagnoses can frequently be obtained using a neural network, but accuracy can be improved with further data accumulation.

Artificial Intelligence↗
PubMed1994

Computerized detection of clustered microcalcifications in digital mammograms using a shift-invariant artificial neural network.

A computer-aided diagnosis (CAD) scheme has been developed in our laboratory for the detection of clustered microcalcifications in digital mammograms. In this study, we apply a shift-invariant neural network to eliminate false-positive detections reported by the CAD scheme. The shift-invariant neural network is a multilayer back-propagation neural network with local, shift-invariant interconnections. The advantage of the shift-invariant neural network is that the result of the network is not dependent on the locations of the clustered microcalcifications in the input layer. The neural network is trained to detect each individual microcalcification in a given region of interest (ROI) reported by the CAD scheme. A ROI is classified as a positive ROI if the total number of microcalcifications detected in the ROI is greater than a certain number. The performance of the shift-invariant neural network was evaluated by means of a jackknife (or holdout) method and ROC analysis using a database of 168 ROIs, as reported by the CAD scheme when applied to 34 mammograms. The analysis yielded an average area under the ROC curve (Az) of 0.91. Approximately 55% of false-positive ROIs were eliminated without any loss of the true-positive ROIs. The result is considerably better than that obtained in our previous study using a conventional three-layer, feed-forward neural network. The effect of the network structure on the performance of the shift-invariant neural network is also studied.

Biophysical Phenomena↗
PubMed1994

Minimization of EOG artefacts from corrupted EEG signals using a neural network approach.

In this paper, we propose a neural network (NN) approach to the enhancement of EEG signals in the presence of EOG artefacts. We recast the EEG enhancement problem into the optimization framework by developing an appropriate cost function. The cost function is nothing but the energy in the enhanced EEG signal obtained through a nonlinear filter formulation, unlike the conventionally-used linear filter formulation. The minimization property of feedback-type neural networks is exploited to solve this problem. An analysis has been performed to characterize the stationary points of the suggested energy function. The hardware set-up of the developed neural network has also been derived. The optimum nonlinear filter coefficients obtained from this minimization algorithm are used to estimate the EOG artefact which is then subtracted from the corrupted EEG signal, sample by sample, to get the artefact minimized signal. The time plots as the LP spectrum show that the proposed method is very effective. Thus the power and efficacy of the NN approach have been exploited for the purpose of minimizing EOG artefacts from corrupted EEG signals.

Algorithms↗
PubMed2006

Percolation in living neural networks.

We study living neural networks by measuring the neurons' response to a global electrical stimulation. Neural connectivity is lowered by reducing the synaptic strength, chemically blocking neurotransmitter receptors. We use a graph-theoretic approach to show that the connectivity undergoes a percolation transition. This occurs as the giant component disintegrates, characterized by a power law with an exponent beta approximately or = 0.65. Beta is independent of the balance between excitatory and inhibitory neurons and indicates that the degree distribution is Gaussian rather than scale free.

Animals↗
PubMed2002

Optimization of artificial neural networks used for retention modelling in ion chromatography.

The aim of this work is the development of an artificial neural network model, which can be generalized and used in a variety of applications for retention modelling in ion chromatography. Influences of eluent flow-rate and concentration of eluent anion (OH-) on separation of seven inorganic anions (fluoride, chloride, nitrite, sulfate, bromide, nitrate, and phosphate) were investigated. Parallel prediction of retention times of seven inorganic anions by using one artificial neural network was applied. MATLAB Neural Networks ToolBox was not adequate for application to retention modelling in this particular case. Therefore the authors adopted it for retention modelling by programming in MATLAB metalanguage. The following routines were written; the division of experimental data set on training and test set; selection of data for training and test set; Dixon's outlier test; retraining procedure routine; calculations of relative error. A three-layer feed forward neural network trained with a Levenberg-Marquardt batch error back propagation algorithm has been used to model ion chromatographic retention mechanisms. The advantage of applied batch training methodology is the significant increase in speed of calculation of algorithms in comparison with delta rule training methodology. The technique of experimental data selection for training set was used allowing improvement of artificial neural network prediction power. Experimental design space was divided into 8-32 subspaces depending on number of experimental data points used for training set. The number of hidden layer nodes, the number of iteration steps and the number of experimental data points used for training set were optimized. This study presents the very fast (300 iteration steps) and very accurate (relative error of 0.88%) retention model, obtained by using a small amount of experimental data (16 experimental data points in training set). This indicates that the method of choice for retention modelling in ion chromatography is the artificial neural network.

Chromatography, Liquid↗
PubMed2006

Associative memory design for 256 gray-level images using a multilayer neural network.

A design procedure is presented for neural associative memories storing gray-scale images. It is an evolution of a previous work based on the decomposition of the image with 2L gray levels into L binary patterns, stored in L uncoupled neural networks. In this letter, an L-layer neural network is proposed with both intralayer and interlayer connections. The connections between different layers introduce interactions among all the neurons, increasing the recall performance with respect to the uncoupled case. In particular, the proposed network can store images with the commonly used number of 256 gray levels instead of 16, as in the previous approach.

Algorithms↗
PubMed2002

Prediction of electrophoretic mobilities of alkyl- and alkenylpyridines in capillary electrophoresis using artificial neural networks.

The electrophoretic mobilities of 31 isomeric alkyl- and alkenylpyridines in capillary electrophoresis were predicted using an artificial neural network (ANN). The multiple linear regression (MLR) technique was used to select the descriptors as inputs for the artificial neural network. The neural network is a fully connected back-propagation model with a 3-6-1 architecture. The results obtained using the neural network were compared with those obtained using the MLR technique. Standard error of training and standard error of prediction were 6.28 and 5.11%, respectively, for the MLR model and 1.03 and 1.20%, respectively, for the ANN model. Two geometric parameters and one electronic descriptor that were used as inputs in the ANN are able to distinguish between the isomers.

Electrophoresis, Capillary↗
PubMedDate not supplied

Evaluation of an artificial neural network to predict urea nitrogen appearance for critically ill multiple-trauma patients.

BACKGROUND: Computer-based simulated biologic neural network models have made significant strides in clinical medicine. METHODS: To determine the predictive performance of a conventional regression model and an artificial neural network for estimating urea nitrogen appearance (UNA) during critical illness, 125 adult patients admitted to the trauma intensive care unit who required specialized nutrition support were studied. The first 100 consecutive patients were used to develop the 2 models. The first model used stepwise multivariate regression analysis. The second model entailed the use of a feeding-forward, back-propagation, supervised neural network. Bias and precision of both methods were evaluated in 25 separate patients. RESULTS: Multivariate regression analysis revealed a significant highly correlative relationship (r(2) = .918, p < or = .01): Predicted UNA (g/d) = (0.29 x WT) + (1.20 x WBC) + (0.44 x SUN) with WT as current body weight in kg, WBC as white blood cell count in cells/mm(3), and SUN as serum urea nitrogen concentration (mg/dL). The regression method was biased toward overestimating measured UNA, whereas the neural network was unbiased. Precision (95% confidence interval) of the neural network was significantly better than the regression (3.3-7.2 g vs 7.3-11.6 g, respectively, p < .01). Regression analysis successfully predicted UNA within 3 g of measured UNA in 16% (4 of 25) of patients, whereas the neural network successfully predicted UNA in 44% (11 out of 25) of patients (p < .06). CONCLUSIONS: These preliminary data indicate that use of an artificial neural network may be superior to conventional regression modeling techniques for estimating UNA in critically ill adult multiple-trauma patients receiving specialized nutrition support.

Adult↗
PubMed1996

On the analysis of the inverse problem of metabolic pathways using artificial neural networks.

Here we develop the use of artificial neural networks for solving the inverse metabolic problem, in other words, given a set of steady-state metabolite levels and fluxes in a pathway of known structure to obtain the parameters of the system, in this case the enzymatic limiting rate and Michaelis constants. This requires two main procedures: first the development of a computer program with which one can model metabolism in the forward direction (i.e. given the internal and parameters to determine the steady-state fluxes and metabolite concentrations), and second, given arrays of associated parameters and variables thereby obtained, to exploit artificial neural networks to form a model capable of obtaining the parameters from the variables. We studied 2-step pathways exhibiting first-order kinetics, 2-step pathways exhibiting reversible Michaelis-Menten kinetics and then 3-step pathways (again exhibiting reversible Michaelis-Menten kinetics), modelled using the program Gepasi. Whilst it was fairly easy for the networks to learn most of the parameters in the 2-step pathway, it was found helpful for the Michaelis-Menten case to vary the concentration of the starting pathway substrate for each set of internal parameters, and to train separate networks for each parameter. Some parameters were much easier to learn than others, reverse K(m) and V(max) values normally being the most difficult. For the 3-step pathway learning sometimes required as much as 3 days, and occasionally convergence was not obtained. Overall, neural networks of the present type, with fully interconnected feedforward architectures and trained according to the backpropagation algorithm, scaled poorly as the problem size was increased.

Algorithms↗
PubMed2001

Understanding neural networks using regression trees: an application to multiple myeloma survival data.

Neural networks are becoming very popular tools for analysing data. It is however quite difficult to understand the neural network output in terms of the original covariates or input variables. In this paper we provide, using readily available software, an easy way of understanding the output of the neural network using regression trees. We focus on the problem in the context of censored survival data for patients with multiple myeloma, where identifying groups of patients with different prognosis is an important aspect of clinical studies. The use of regression trees to help understand neural networks can be easily applied to uncensored situations.

Age Factors↗