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 793 records · Page 44Linked to original sources
PubMed2000

A neural network approach for insulin regime and dose adjustment in type 1 diabetes.

BACKGROUND: A decision support system based on a neural network approach is proposed to advise on insulin regime and dose adjustment for type 1 diabetes patients. METHOD: The system consists of two feed-forward neural networks, trained with the back-propagation algorithm with momentum and adaptive learning rate. The input to the system consists of patient's glucose levels, insulin intake, and observed hypoglycemia symptoms during a short time period. The output of the first neural network provides the insulin regime, which is applied as input to the second neural network to estimate the appropriate insulin doses for a short time period. RESULTS: The system's ability in order to recommend on insulin regime is excellent, while its performance in adjusting the insulin dosages for a specific patient is highly dependent on the data set used during the training procedure. CONCLUSIONS: Despite the limitations of computer-based approaches, this study shows that artificial neural networks can assist diabetes patients in insulin adjustment.

Blood Glucose↗
PubMed1996

Automatic detection of diabetic retinopathy using an artificial neural network: a screening tool.

AIMS: To determine if neural networks can detect diabetic features in fundus images and compare the network against an ophthalmologist screening a set of fundus images. METHODS: 147 diabetic and 32 normal images were captured from a fundus camera, stored on computer, and analysed using a back propagation neural network. The network was trained to recognise features in the retinal image. The effects of digital filtering techniques and different network variables were assessed. 200 diabetic and 101 normal images were then randomised and used to evaluate the network's performance for the detection of diabetic retinopathy against an ophthalmologist. RESULTS: Detection rates for the recognition of vessels, exudates, and haemorrhages were 91.7%, 93.1%, and 73.8% respectively. When compared with the results of the ophthalmologist, the network achieved a sensitivity of 88.4% and a specificity of 83.5% for the detection of diabetic retinopathy. CONCLUSIONS: Detection of vessels, exudates, and haemorrhages was possible, with success rates dependent upon preprocessing and the number of images used in training. When compared with the ophthalmologist, the network achieved good accuracy for the detection of diabetic retinopathy. The system could be used as an aid to the screening of diabetic patients for retinopathy.

Diabetic Retinopathy↗
PubMed2001

The ability of the GDx nerve fibre analyser neural network to diagnose glaucoma.

PURPOSE: To evaluate the neural network used by the GDx in a group of normal subjects, patients with ocular hypertension (OHT) and patients with normal-pressure glaucoma (NPG). METHODS: The GDx neural network produces a "number" that indicates the likelihood that glaucoma is present. This number was compared in three groups representing different stages of health and disease, namely, normal controls (n = 101), OHT (n = 102) and NPG (105). The GDx number's ability to differentiate between normal and glaucoma individuals was then investigated. We also studied the relationship between the GDx number and retinal nerve fibre layer (RNFL) average thickness and visual field status to examine how well the GDx number reflects disease severity. RESULTS: The GDx number was significantly different among the groups (P < 0.01); it was highest in NPG and lowest in normal controls. The GDx number differentiated between glaucoma and normal with sensitivity of 92.3% and specificity of 96%. When combined with the parameter of RNFL average thickness, sensitivity and specificity were 88.5% and 100% respectively. In NPG a significant correlation was found between the GDx number and RNFL average thickness(rho = -0.88, P < 0.001) and visual field mean deviation (rho = -0.64, P < 0.001). CONCLUSION: The GDx number is able to differentiate between groups of normal, OHT and NPG subjects. Its close relationship with RNFL average thickness and visual field status in glaucoma indicates that it is able to reflect disease severity. Furthermore, its measured ability to distinguish between normal individuals and those with glaucoma demonstrates potential for use in glaucoma diagnosis.

Adult↗
PubMed2000

Using an artificial neural network to detect activations during ventricular fibrillation.

Ventricular fibrillation is a cardiac arrhythmia that can result in sudden death. Understanding and treatment of this disorder would be improved if patterns of electrical activation could be accurately identified and studied during fibrillation. A feedforward artificial neural network using backpropagation was trained with the Rule-Based Method and the Current Source Density Method to identify cardiac tissue activation during fibrillation. Another feedforward artificial neural network that used backpropagation was trained with data preprocessed by those methods and the Transmembrane Current Method. Staged training, a new method that uses different sets of training examples in different stages, was used to improve the ability of the artificial neural networks to detect activation. Both artificial neural networks were able to correctly classify more than 92% of new test examples. The performance of both artificial neural networks improved when staged training was used. Thus, artificial neural networks may beuseful for identifying activation during ventricular fibrillation.

Diagnosis, Computer-Assisted↗
PubMed1998

Neural network subtyping of depression.

OBJECTIVE: To examine the applicability of a neural network classification strategy to examine the independent contribution of psychomotor disturbance (PMD) and endogeneity symptoms to the DSM-III-R definition of melancholia. METHOD: We studied 407 depressed patients with the clinical dataset comprising 17 endogeneity symptoms and the 18-item CORE measure of behaviourally rated PMD. A multilayer perception neural network was used to fit non-linear models of varying complexity. A linear discriminant function analysis was also used to generate a model for comparison with the non-linear models. RESULTS: Models (linear and non-linear) using PMD items only and endogeneity symptoms only had similar rates of successful classification, while non-linear models combining both PMD and symptoms scores achieved the best classifications. CONCLUSIONS: Our current non-linear model was superior to a linear analysis, a finding which may have wider application to psychiatric classification. Our non-linear analysis of depressive subtypes supports the binary view that melancholic and non-melancholic depression are separate clinical disorders rather than different forms of the same entity. This study illustrates how non-linear modelling with neural networks is a potentially fruitful approach to the study of the diagnostic taxonomy of psychiatric disorders and to clinical decision-making.

Behavioral Symptoms↗
PubMed1999

Model selection in neural networks.

In this article, we examine how model selection in neural networks can be guided by statistical procedures such as hypothesis tests, information criteria and cross validation. The application of these methods in neural network models is discussed, paying attention especially to the identification problems encountered. We then propose five specification strategies based on different statistical procedures and compare them in a simulation study. As the results of the study are promising, it is suggested that a statistical analysis should become an integral part of neural network modeling.

Journal Article↗
PubMed1996

Binary-oscillator networks: bridging a gap between experimental and abstract modeling of neural networks.

This paper proposes a simplified oscillator model, called binary-oscillator, and develops a class of neural network models having binary-oscillators as basic units. The binary-oscillator has a binary dynamic variable v = +/- 1 modeling the "membrane potential" of a neuron, and due to the presence of a "slow current" (as in a classical relaxation-oscillator) it can oscillate between two states. The purpose of the simplification is to enable abstract algorithmic study on the dynamics of oscillator networks. A binary-oscillator network is formally analogous to a system of stochastic binary spins (atomic magnets) in statistical mechanics.

Algorithms↗
PubMed1998

Use of an artificial neural network to predict length of stay in acute pancreatitis.

Length of stay (LOS) predictions in acute pancreatitis could be used to stratify patients with severe acute pancreatitis, make treatment and resource allocation decisions, and for quality assurance. Artificial neural networks have been used to predict LOS in other conditions but not acute pancreatitis. The hypothesis of this study was that a neural network could predict LOS in patients with acute pancreatitis. The medical records of 195 patients admitted with acute pancreatitis were reviewed. A backpropagation neural network was developed to predict LOS >7 days. The network was trained on 156 randomly selected cases and tested on the remaining 39 cases. The neural network had the highest sensitivity (75%) for predicting LOS >7 days. Ranson criteria had the highest specificity (94%) for making this prediction. All methods incorrectly predicted LOS in two patients with severe acute pancreatitis who died early in their hospital course. An artificial neural network can predict LOS >7 days. The network and traditional prognostic indices were least accurate for predicting LOS in patients with severe acute pancreatitis who died early in their hospital course. The neural network has the advantage of making this prediction using admission data.

APACHE↗
PubMed2005

Optimization via intermittency with a self-organizing neural network.

One of the major obstacles in using neural networks to solve combinatorial optimization problems is the convergence toward one of the many local minima instead of the global minima. In this letter, we propose a technique that enables a self-organizing neural network to escape from local minima by virtue of the intermittency phenomenon. It gives rise to novel search dynamics that allow the system to visit multiple global minima as meta-stable states. Numerical experiments performed suggest that the phenomenon is a combined effect of Kohonen-type competitive learning and the iterated softmax function operating near bifurcation. The resultant intermittent search exhibits fractal characteristics when the optimization performance is at its peak in the form of 1/f signals in the time evolution of the cost, as well as power law distributions in the meta-stable solution states. TheN-Queens problem is used as an example to illustrate the meta-stable convergence process that sequentially generates, in a single run, 92 solutions to the 8-Queens problem and 4024 solutions to the 17-Queens problem.

Algorithms↗
PubMed1997

Neural network classification and quantification of organic vapors based on fluorescence data from a fiber-optic sensor array.

Computational neural networks have been developed to classify and quantify nine organic vapors. The neural network analyses used data that consisted of the change in fluorescence from a sensor array that consisted of 19 fiber optics with immobilized dye in polymer matrices. Plots of change in fluorescence intensity versus time were measured as pulses of analyte were presented to the sensor array. Descriptors were calculated from the intensity vs time plots, and they were used to build neural network models that accurately classified and quantified each of the nine analytes. Most of the data were used to train the neural networks (training set members), some were used to assist termination of training (cross-validation set members), and some were used to validate the models (prediction set members). Classification rates approaching 100% were achieved for the training set data, and 90% of the members in the prediction set were correctly classified. In addition, 97% of the prediction set observations were assigned a correct relative concentration.

Acetates↗
PubMed2001

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

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

Autopsy↗
PubMed2002

Comparing neural networks and linear discriminant functions for glaucoma detection using confocal scanning laser ophthalmoscopy of the optic disc.

PURPOSE: To determine whether neural network techniques can improve differentiation between glaucomatous and nonglaucomatous eyes, using the optic disc topography parameters of the Heidelberg Retina Tomograph (HRT; Heidelberg Engineering, Heidelberg, Germany). METHODS: With the HRT, one eye was imaged from each of 108 patients with glaucoma (defined as having repeatable visual field defects with standard automated perimetry) and 189 subjects without glaucoma (no visual field defects with healthy-appearing optic disc and retinal nerve fiber layer on clinical examination) and the optic nerve topography was defined by 17 global and 66 regional HRT parameters. With all the HRT parameters used as input, receiver operating characteristic (ROC) curves were generated for the classification of eyes, by three neural network techniques: linear and Gaussian support vector machines (SVM linear and SVM Gaussian, respectively) and a multilayer perceptron (MLP), as well as four previously proposed linear discriminant functions (LDFs) and one LDF developed on the current data with all HRT parameters used as input. RESULTS: The areas under the ROC curves for SVM linear and SVM Gaussian were 0.938 and 0.945, respectively; for MLP, 0.941; for the current LDF, 0.906; and for the best previously proposed LDF, 0.890. With the use of forward selection and backward elimination optimization techniques, the areas under the ROC curves for SVM Gaussian and the current LDF were increased to approximately 0.96. CONCLUSIONS: Trained neural networks, with global and regional HRT parameters used as input, improve on previously proposed HRT parameter-based LDFs for discriminating between glaucomatous and nonglaucomatous eyes. The performance of both neural networks and LDFs can be improved with optimization of the features in the input. Neural network analyses show promise for increasing diagnostic accuracy of tests for glaucoma.

Discriminant Analysis↗
PubMed2000

[Study on the determination of contents of Zn, Cu and Co by using artifical neural network and ultraviolet spectrum].

By means of artificial neural network and back-propagation train algorithm, the three-component metal coordinate compounds of PAR-Zn, Cu, Co were determined simultaneously, in which the spectra overlapped. In 580-440 nm, the absorbance(A) at 14 wavelength were taken as character parameter of artificial neural network, and samples were arranged by method of uniformity design. The mean recovery of Zn, Cu, Co were 95.22%, 95.98% and 100.5%. The experiment results show that the determination is accurate, and the method has good performance.

Cobalt↗
PubMed1991

Langevin machine: a neural network based on stochastically justifiable sigmoidal function.

In neural networks the activation process controls the output as a nonlinear function of the input; and, this output remains bounded between limits as decided by a logistic function known as the sigmoid (S-shaped). Presently, by applying the considerations of Maxwell-Boltzmann statistics, the Langevin function is shown as the appropriate and justifiable sigmoid (instead of the conventional hyperbolic tangent function) to depict the bipolar nonlinear logic-operation enunciated by the collective stochastical response of artificial neurons under activation. That is, the graded response of a large network of 'neurons' such as Hopfield's can be stochastically justified via the proposed model. The model is consistent with the established link between the Hopfield model and the statistical mechanics. The Langevin function (in lieu of conventional hyperbolic tangent and/or exponential sigmoids) in determining nonlinear decision boundaries, in characterizing the neural networks by the Langevin machine versus the Boltzmann machine, in sharpening and annealing schedules and in the optimization of nonlinear detector performance are discussed.

Animals↗
PubMed1993

Prediction of valve-related complications for artificial heart valves using adaptive neural networks: a preliminary study.

A novel approach to the prediction of valve-related complications in patients with implanted artificial heart valves is discussed. Adaptive artificial neural networks were used to identify patients at high risk of valve-related events based on preoperative data. Data from a clinical trial on 789 subjects with Carpentier-Edwards pericardial bioprostheses were used. Patients' records were divided into two groups, one of which was used for training the neural network and the other for testing the trained network and determining error rates. Patient information such as age, sex, NYHA class and anticoagulation therapy, as well as valve information such as size and the date of implant, were used as the network inputs. The neural net had a single output variable indicating the risk that an individual patient would develop a valve-related complication resulting in death. The results show that a trained neural network was able to predict valve-related deaths in the specified time interval of 1981-1991 with a high degree of accuracy. The neural network was also successful in classifying patients into high and low risk categories.

Animals↗
PubMed2004

Neural network analysis of anal sphincter repair.

PURPOSE: Prediction of success after anterior sphincter repair for incontinence is difficult. Standard multivariate analysis techniques have only 75 to 80 percent accuracy. Artificial intelligence, including artificial neural networks, has been used in the analysis of complex clinical data and has proved to be successful in predicting the outcome of other surgical procedures. Using a neural network algorithm, we have assessed the probability of success after anterior sphincter repair. METHODS: Prospective anorectal physiology data of 72 patients undergoing anterior sphincter repair was collected between 1995 and 1999. Complete data sets of 75 percent of the series were used to train an artificial neural network; the remaining 25 percent were used for data validation. The output was continence grading, ranging from 0 to 4 (worse to continent). RESULTS: The outcome at 3, 6, and 12 months postoperatively was obtained and assessed. The best correlation between actual data value and artificial neural network value was found at 12 months (r = 0.931; P = 0.0001). Clear correlations also were found at three months (r = 0.898; P = 0.0001) and six months (r = 0.742; P = 0.002). Results of applying a net to details excluding pudendal nerve latency were poor. CONCLUSIONS: Artificial neural networks are more accurate (93 percent correlation) than standard statistics (75 percent) when applied to the prediction of outcome after anterior sphincter repair. This assessment also confirms the usefulness of pudendal latency in the prediction of anterior sphincter repair outcome. The results obtained highlight the obvious usefulness of artificial neural networks, which could now be used in a prospective evaluation for application of the technique.

Anal Canal↗
PubMed2006

Using neural networks and just nine patient-reportable factors of screen for AMI.

The study investigated the effect of different input selections on the performance of artificial neural networks in screening for acute myocardial infarction (AMI) in Malaysian patients complaining of chest pain. We used hospital data to create neural networks with four input selections and used these to diagnose AMI. A 10-fold cross-validation and committee approach was used. All the neural networks using various input selections outperformed a multiple logistic regression model, although the difference was not statistically significant. The neural networks achieved an area under the ROC curve of 0.792 using nine inputs, whereas multiple logistic regression achieved 0.739 using 64 inputs. Sensitivity levels of over 90 per cent were achieved using low output threshold levels. Specificity levels of over 90 per cent were achieved using threshold levels of 0.4-0.5. Thus neural networks can perform as well as multiple logistic regression models even when using far fewer inputs.

Acute Disease↗