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 397 records · Page 22Linked to original sources
PubMed1997

Application of neural networks to population pharmacokinetic data analysis.

This research examined the applicability of using a neural network approach to analyze population pharmacokinetic data. Such data were collected retrospectively from pediatric patients who had received tobramycin for the treatment of bacterial infection. The information collected included patient-related demographic variables (age, weight, gender, and other underlying illness), the individual's dosing regimens (dose and dosing interval), time of blood drawn, and the resulting tobramycin concentration. Neural networks were trained with this information to capture the relationships between the plasma tobramycin levels and the following factors: patient-related demographic factors, dosing regimens, and time of blood drawn. The data were also analyzed using a standard population pharmacokinetic modeling program, NON-MEM. The observed vs predicted concentration relationships obtained from the neural network approach were similar to those from NONMEM. The residuals of the predictions from neural network analyses showed a positive correlation with that from NONMEM. Average absolute errors were 33.9 and 37.3% for neural networks and 39.9% for NONMEM. Average prediction errors were found to be 2.59 and -5.01% for neural networks and 17.7% for NONMEM. We concluded that neural networks were capable of capturing the relationships between plasma drug levels and patient-related prognostic factors from routinely collected sparse within-patient pharmacokinetic data. Neural networks can therefore be considered to have potential to become a useful analytical tool for population pharmacokinetic data analysis.

Anti-Bacterial Agents↗
PubMed2002

Cultured rat hippocampal neural progenitors generate spontaneously active neural networks.

We previously demonstrated that the neural cell adhesion molecule (N-CAM) inhibited the proliferation of cultured rat hippocampal progenitor cells and increased the number of neurons generated. We demonstrate here that the continued presence of fibroblast growth factor 2 along with N-CAM or brain-derived neurotrophic factor over 12 days of culture greatly increased the number of both progenitors and neurons. These progenitor-derived neurons expressed neurotransmitters, neurotransmitter receptors, and synaptic proteins in vitro consistent with those expressed in the mature hippocampus. Progenitor cells cultured on microelectrode plates formed elaborate neural networks that exhibited spontaneously generated action potentials after 21 days. This activity was observed only in cultures grown in the presence of fibroblast growth factor 2 and either N-CAM or brain-derived neurotrophic factor. Analysis of neuronal activity after various pharmacological treatments indicated that the networks formed functional GABAergic and glutamatergic synapses. We conclude that mitogenic growth factors can synergize with N-CAM or neurotrophins to generate spontaneously active neural networks from neural progenitors.

Action Potentials↗
PubMedDate not supplied

Feasibility of using neural networks to estimate minimum tumour temperature and perfusion values.

We examine the ability of neural networks to estimate the tissue perfusion values present and the minimum temperature in numerically calculated (Pennes, Bioheat Transfer Equation) steady-state hyperthermia temperature fields based on a limited number of measured temperatures within this field A hierarchical system of neural networks consisting of a first layer of pattern recognizing neural networks and a second layer of hypersurface reconstructing neural networks is shown to be capable of estimating these variables within a selected error tolerance. The results indicate that estimating the minimum tumour temperature directly with the system of neural networks may be more effective than using the indirect method of numerically recreating a temperature field with perfusion estimates and then obtaining the minimum tumour temperature from the estimated temperature field. Additional results indicate that if the locations of the measured temperatures within the temperature field are selected appropriately, the hierarchical system of neural networks can tolerate a moderate level of model mismatch. This model mismatch can come from errors in modelling the tumour boundaries, the sensor locations, or the magnitude of the power deposition. This paper is not intended to assess or demonstrate clinical applicability but to be a first step in investigating the feasibility of neural networks for parameter estimation related to hyperthermia studies.

Computer Simulation↗
PubMed1994

Prediction of breast cancer malignancy using an artificial neural network.

BACKGROUND: An artificial neural network (ANN) was developed to predict breast cancer from mammographic findings. This network was evaluated in a retrospective study. METHODS: For a set of patients who were scheduled for biopsy, radiologists interpreted the mammograms and provided data on eight mammographic findings as part of the standard mammographic workup. These findings were encoded as features for an ANN. Results of biopsies were taken as truth in the diagnosis of malignancy. The ANN was trained and evaluated using a jackknife sampling on a set of 260 patient records. Performance of the network was evaluated in terms of sensitivity and specificity over a range of decision thresholds and was expressed as a receiver operating characteristic curve. RESULTS: The ANN performed more accurately than the radiologists (P < 0.08) with a relative sensitivity of 1.0 and specificity of 0.59. CONCLUSIONS: An ANN can be trained to predict malignancy from mammographic findings with a high degree of accuracy.

Algorithms↗
PubMed1999

Neural networks in outcomes research.

Neural networks (NNs) are being used in the areas of prediction and classification of outcomes in medicine--areas in which regression models have traditionally been used. In this report, we summarize the steps in developing and testing an NN. Through applications in clinical research, we present several examples of NN development and evaluation. Through these applications we show that the performance of the NNs matched or exceeded the performance of traditional methods. We then discuss the advantages and disadvantages of NN models in comparison to traditional regression methods.

Health Services Research↗
PubMed1993

Kakadu--a low power analogue neural network classifier.

An analogue neural network VLSI chip designed for low power operation is presented. This chip consists of 84 synapse elements arranged as arrays of size 10 x 6 and 6 x 4 and was fabricated using a standard 1.2 micron double metal single poly CMOS process. The synapses are digitally programmable and static weight storage is provided. The chip has a typical power consumption of tens of microwatts. It has been successfully trained and tested on a range of classification problems including 4-bit parity, character recognition and morphological-based classification of intracardiac electrogram signals.

Algorithms↗
PubMed1997

Pharyngeal wall vibration detection using an artificial neural network.

An artificial-neural-network-based detector of pharyngeal wall vibration (PWV) is presented. PWV signals the imminent occurrence of obstructive sleep apnoea (OSA) in adults who suffer from OSA syndrome. Automated detection of PWV is very important in enhancing continuous positive airway pressure (CPAP) therapy by allowing automatic adjustment of the applied airway pressure by a procedure called automatic positive airway pressure (APAP) therapy. A network with 15 inputs, one output, and two hidden layers, each with two Adaline-nodes, is used as part of a PWV detection scheme. The network is initially trained using nasal mask pressure data from five positively diagnosed OSA patients. The performance of the ANN-based detector is evaluated using data from five different OSA patients. The results show that on the average it correctly detects the presence of PWV events at a rate of approximately 92% and correctly distinguishes normal breaths approximately 98% of the time. Further, the ANN-based detector accuracy is not affected by the pressure level required for therapy.

Adult↗
PubMed2005

Prediction of toxicity using a novel RBF neural network training methodology.

A neural network methodology based on the radial basis function (RBF) architecture is introduced in order to establish quantitative structure-toxicity relationship models for the prediction of toxicity. The dataset used consists of 221 phenols and their corresponding toxicity values to Tetrahymena pyriformis. Physicochemical parameters and molecular descriptors are used to provide input information to the models. The performance and predictive abilities of the RBF models are compared to standard multiple linear regression (MLR) models. The leave-one-out cross validation procedure and validation through an external test set produce statistically significant R2 and RMS values for the RBF models, which prove considerably more accurate than the MLR models. [Figure: see text].

Animals↗
PubMedDate not supplied

Quantitative structure-activity relationships in carboquinones and benzodiazepines using counter-propagation neural networks.

Counter-propagation neural networks are used to model and predict activities of carboquinones and of benzodiazepines from physicochemical parameters. For carboquinones, networks with one hidden layer processing element (PE) for each compound achieved significantly better training set RMSE values than corresponding back-propagation and multiregression results and test set RMSE values as good or slightly worse than back-propagation. Test set results improved by 10-15% using networks with fewer hidden layer PEs than carboquinones; the smallest test set RMSE values are between 0 and 10% better than back-propagation values, about 1.3 times greater than corresponding training set values, and occur when there are about as many competitive layer PEs as there are compounds in the data set. Training set RMSE values increase with decreasing number of competitive layer PEs and approach those of test sets. Both counter-propagation and back-propagation networks, however, have worse predictive capability than multiregression. For benzodiazepines, networks with one hidden layer PE for each compound achieved significantly better training set RMSE values than back-propagation and multiregression results and test set RMSE values slightly worse than back-propagation. Test set results improved by 10-15% using fewer hidden layer PEs than benzodiazepines; the smallest test set RMSE values are 0-10% better than back-propagation values, about 1.3 times greater than training set values, and occur when there are about half as many competitive layer PEs as there are compounds in the data set. Training set RMSE values increase with decreasing number of competitive layer PEs and approach those of test sets. Counter-propagation, back-propagation, and multiregression all have similar predictive capabilities.

Benzodiazepines↗
PubMed2005

The confusion effect in predatory neural networks.

A simple artificial neural network model of image reconstruction in sensory maps is presented to explain the difficulty predators experience in targeting prey in large groups (the confusion effect). Networks are trained to reconstruct multiple randomly conformed "retinal" images of prey groups in an internal spatial map of their immediate environment. They are then used to simulate prey targeting by predators on groups of specific conformation. Networks trained with the biologically plausible associative reward-penalty method produce a more realistic model of the confusion effect than those trained with the popular but biologically implausible backpropagation method. The associative reward-penalty model makes the novel prediction that the accuracy-group size relationship is U shaped, and this prediction is confirmed by empirical data gathered from interactive computer simulation experiments with humans as "predators." The model further predicts all factors known from previous empirical work (and most factors suspected) to alleviate the confusion effect: increased relative intensity of the target object, heterogeneity of group composition, and isolation of the target. Interestingly, group compaction per se is not predicted to worsen predator confusion. This study indicates that the relatively simple, nonattentional mechanism of information degradation in the sensory mapping process is potentially important in generating the confusion effect.

Animals↗
PubMed1996

Orange juice classification with a biologically based neural network.

Dystal, an artificial neural network, was used to classify orange juice products. Nine varieties of oranges collected from six geographical regions were processed into single-strength, reconstituted or frozen concentrated orange juice. The data set represented 240 authentic and 173 adulterated samples of juices; 16 variables [8 flavone and flavanone glycoside concentrations measured by high-performance liquid chromatography (HPLC) and 8 trace element concentrations measured by inductively coupled plasma spectroscopy] were selected to characterize each juice and were used as input to Dystal. Dystal correctly classified 89.8% of the juices as authentic or adulterated. Classification performance increased monotonically as the percentage of pulpwash in the sample increased. Dystal correctly identified 92.5% of the juices by variety (Valencia vs non-Valencia).

Algorithms↗
PubMed2004

A one-layer recurrent neural network for support vector machine learning.

This paper presents a one-layer recurrent neural network for support vector machine (SVM) learning in pattern classification and regression. The SVM learning problem is first converted into an equivalent formulation, and then a one-layer recurrent neural network for SVM learning is proposed. The proposed neural network is guaranteed to obtain the optimal solution of support vector classification and regression. Compared with the existing two-layer neural network for the SVM classification, the proposed neural network has a low complexity for implementation. Moreover, the proposed neural network can converge exponentially to the optimal solution of SVM learning. The rate of the exponential convergence can be made arbitrarily high by simply turning up a scaling parameter. Simulation examples based on benchmark problems are discussed to show the good performance of the proposed neural network for SVM learning.

Journal Article↗
PubMed2004

Detecting acute myocardial infarction in the 12-lead ECG using Hermite expansions and neural networks.

We use artificial neural networks (ANNs) to detect signs of acute myocardial infarction (AMI) in ECGs. The 12-lead ECG is decomposed into Hermite basis functions, and the resulting coefficients are used as inputs to the ANNs. Furthermore, we present a case-based method that qualitatively explains the operation of the ANNs, by showing regions of each ECG critical for ANN response. Key ingredients in this method are: (i) a cost function used to find local ECG perturbations leading to the largest possible change in ANN output and (ii) a minimization scheme for this cost function using mean field annealing. Our approach was tested on 2238 ECGs recorded at an emergency department. The obtained ROC areas for ANNs trained with the Hermite representation and standard ECG measurements were 83.4 and 84.3% (P=0.4), respectively. We believe that the proposed method has potential as a decision support system that can provide good advice for diagnosis, as well as providing the physician with insight into the reason underlying the advice.

Acute Disease↗
PubMed1996

Autonomous design of artificial neural networks by Neurex.

Artificial neural networks (ANN) have been demonstrated to be increasingly more useful for complex problems difficult to solve with conventional methods. With their learning abilities, they avoid having to develop a mathematical model or acquiring the appropriate knowledge to solve a task. The difficulty now lies in the ANN design process. A lot of choices must be made to design an ANN, and there are no available design rules to make these choices directly for a particular problem. Therefore, the design for an ANN demands a certain number of iterations, mainly guided by the expertise and the intuition of the developer. To automate the ANN design process, we have developed Neurex, composed of an expert system and an ANN simulator. Neurex autonomously guides the iterative ANN design process. Its structure tries to reproduce the design steps done by a human expert in conceiving an ANN. As a whole, the Neurex structure serves as a framework to implement this expertise for different learning paradigms. This article presents the system's general characteristics and its use in designing ANN using the standard backpropagation learning law.

Computer Simulation↗
PubMed1998

Predicting lettuce canopy photosynthesis with statistical and neural network models.

An artificial neural network (NN) and a statistical regression model were developed to predict canopy photosynthetic rates (Pn) for 'Waldman's Green' leaf lettuce (Latuca sativa L.). All data used to develop and test the models were collected for crop stands grown hydroponically and under controlled-environment conditions. In the NN and regression models, canopy Pn was predicted as a function of three independent variables: shootzone CO2 concentration (600 to 1500 micromoles mol-1), photosynthetic photon flux (PPF) (600 to 1100 micromoles m-2 s-1), and canopy age (10 to 20 days after planting). The models were used to determine the combinations of CO2 and PPF setpoints required each day to maintain maximum canopy Pn. The statistical model (a third-order polynomial) predicted Pn more accurately than the simple NN (a three-layer, fully connected net). Over an 11-day validation period, average percent difference between predicted and actual Pn was 12.3% and 24.6% for the statistical and NN models, respectively. Both models lost considerable accuracy when used to determine relatively long-range Pn predictions (> or = 6 days into the future).

Carbon Dioxide↗
PubMed2003

Prediction of skin penetration using artificial neural network (ANN) modeling.

Artificial neural network (ANN) analysis was used to predict the skin permeability of selected xenobiotics. Permeability coefficients (log k(p)) were obtained from various literature sources. A previously reported equation, which was shown to be useful in the prediction of skin permeability, uses the partial charges of the penetrants, their molecular weight, and their calculated octanol water partition coefficient (log K(oct)). The equation was used to predict the skin permeability for the set of 40 compounds (r(2) = 0.672). A successful ANN was developed and the ANN produced log k(p) values that correlated well with the experimental ones(r(2) = 0.997). The penetration properties of a selection of compounds through human skin that have not been previously investigated, etodolac, famotidine, nimesulide, nizatidine, ranitidine, were investigated. Their permeability coefficients were determined. It was then possible to compare the experimental data with that predicted using the partial charge equation and the trained ANN. ANN modeling for predicting skin permeability was found to be useful for predicting skin permeability coefficients of compounds. In conclusion, the developed and described ANN model in this publication does not require any experimental parameters; it could potentially provide useful and precise prediction of skin penetration for new drugs or toxic penetrants.

Artificial Intelligence↗
PubMed1995

Classification of arrhythmic events in ambulatory electrocardiogram, using artificial neural networks.

We propose artificial neural networks (ANN) for ambulatory ECG arrhythmic event classification, and we compare them with some traditional classifiers (TC). Among them, the one based on the median method (heuristic algorithm) was chosen and taken as a quality reference in this study, while a back propagation based classifier, designed as an autoassociator for its peculiar capability of rejecting unknown patterns, was examined. Two tests were performed: the first to discriminate normal vs ventricular beats and the second to distinguish among three classes of arrhythmic events. The results show that the ANN approach is more reliable than the traditional classifiers in discriminating among many classes of arrhythmic events: 98% by ANN vs 99% by a TC for correctly classified normal beats, 98% by ANN vs 96% by TC for correctly classified ventricular ectopic beats, 96% by ANN vs 59% by TC for correctly classified supraventricular ectopic beats, and 83% by ANN vs 86% by median method for correctly classified aberrated atrial premature beats. This paper also tackles the problem of the management of classification uncertainty. Two concurrent uncertainty criteria have been introduced, to reduce the classification error of the unknown ventricular and supraventricular arrhythmic beats respectively. The error in ventricular beats case was kept close to 0% in average and for supraventricular beats was kept at 35% in average. So we can state that the ANN approach is powerful in classifying beats represented in the training set and that it manages the uncertainty in such a way as to reduce, in any case, the global error percentage.

Algorithms↗