Search PubMed⌕ Search

SEARCH · Search PubMed

Results for “Neural network”

Search indexed PubMed citations on genomics, clinical trials, systematic reviews and public health. Explore titles, authors and supplied subject terms, then open the PubMed record.

Quote a phrase for an exact phrase match. Source license links do not imply unrestricted reuse.

At least 1,009 records · Page 56Linked to original sources
PubMed2000

Neural network modeling for surgical decisions on traumatic brain injury patients.

Computerized medical decision support systems have been a major research topic in recent years. Intelligent computer programs were implemented to aid physicians and other medical professionals in making difficult medical decisions. This report compares three different mathematical models for building a traumatic brain injury (TBI) medical decision support system (MDSS). These models were developed based on a large TBI patient database. This MDSS accepts a set of patient data such as the types of skull fracture, Glasgow Coma Scale (GCS), episode of convulsion and return the chance that a neurosurgeon would recommend an open-skull surgery for this patient. The three mathematical models described in this report including a logistic regression model, a multi-layer perceptron (MLP) neural network and a radial-basis-function (RBF) neural network. From the 12,640 patients selected from the database. A randomly drawn 9480 cases were used as the training group to develop/train our models. The other 3160 cases were in the validation group which we used to evaluate the performance of these models. We used sensitivity, specificity, areas under receiver-operating characteristics (ROC) curve and calibration curves as the indicator of how accurate these models are in predicting a neurosurgeon's decision on open-skull surgery. The results showed that, assuming equal importance of sensitivity and specificity, the logistic regression model had a (sensitivity, specificity) of (73%, 68%), compared to (80%, 80%) from the RBF model and (88%, 80%) from the MLP model. The resultant areas under ROC curve for logistic regression, RBF and MLP neural networks are 0.761, 0.880 and 0.897, respectively (P < 0.05). Among these models, the logistic regression has noticeably poorer calibration. This study demonstrated the feasibility of applying neural networks as the mechanism for TBI decision support systems based on clinical databases. The results also suggest that neural networks may be a better solution for complex, non-linear medical decision support systems than conventional statistical techniques such as logistic regression.

Adolescent↗
PubMed2002

Neural network predicts sequence of TP53 gene based on DNA chip.

UNLABELLED: We have trained an artificial neural network to predict the sequence of the human TP53 tumor suppressor gene based on a p53 GeneChip. The trained neural network uses as input the fluorescence intensities of DNA hybridized to oligonucleotides on the surface of the chip and makes between zero and four errors in the predicted 1300 bp sequence when tested on wild-type TP53 sequence. AVAILABILITY: The trained neural network is available for academic use by contacting steen@cbs.dtu.dk

Base Sequence↗
PubMed1996

Discrimination of forearm's motions by surface EMG signals using neural network.

We tried to discriminate different forearm's motions by surface EMG signals using neural network. In order to get a higher discrimination rate, the positions of electrodes were improved. We also tried to discriminate similar motions in order to clarify the limitation of the discrimination by surface EMG signals. Two experiments were carried out. One was to discriminate five different motions: grasp, wrist flexion, wrist extension, forearm pronation, and forearm supination (Experiment 1). The other was to discriminate four similar motions which have different quantitative definitions at grasp, wrist flexion/ extension, or forearm pronation/supination (Experiment 2). Four surface electrodes were placed on the skin above the main active muscles: short radial extensor m. of wrist, supinator m., long radial extensor m. of wrist, and ulnar flexor m. of wrist, considering anatomical functions of the forearm's muscles. EMG signals were recorded during 2 sec while the subjects kept the motions. Recorded EMG signals were sampled at 200 msec intervals after full-wave rectifying and low-pass filtering. Therefore, the number of sampling data patterns of EMG signals was 10 for every motion. Three layers of neural network was used for discrimination. The number of units in the input layer is 4, and the number of units in the output layer is 5 or 4. In order to get the best discrimination rate of the motions, we changed the number of units in the hidden layer from 3 to 12. The neural network was trained by the back-propagation algorithm. In Experiment 1, the best average values of discrimination rates under three patterns of EMG signals for each subject were 96.0%, 98.0%, and 87.2% when the numbers of units in the hidden layer were 10, 11, and 3 respectively. In Experiment 2 using original EMG patterns, the best average values of discrimination rates at grasp, extension/flexion, and pronation/supination were 59.5%, 76.0%, and 25.0% respectively. By using normalized EMG patterns, these were 40.0%, 84.8%, and 55.5% respectively.

Adult↗
PubMed2004

Attractability and location of equilibrium point of cellular neural networks with time-varying delays.

This paper presents new theoretical results on global exponential stability of cellular neural networks with time-varying delays. The stability conditions depend on external inputs, connection weights and delays of cellular neural networks. Using these results, global exponential stability of cellular neural networks can be derived, and the estimate for location of equilibrium point can also be obtained. Finally, the simulating results demonstrate the validity and feasibility of our proposed approach.

Animals↗
PubMed2001

Comparing Bayesian neural network algorithms for classifying segmented outdoor images.

In this paper we investigate the Bayesian training of neural networks for region labelling of segmented outdoor scenes; the data are drawn from the Sowerby Image Database of British Aerospace. Neural networks are trained with two Bayesian methods, (i) the evidence framework of MacKay (1992a,b) and (ii) a Markov Chain Monte Carlo method due to Neal (1996). The performance of the two methods is compared to evaluating the empirical learning curves of neural networks trained with the two methods. We also investigate the use of the Automatic Relevance Determination method for input feature selection.

Algorithms↗
PubMed1994

Generation of associative processes in a neural network with realistic features of architecture and units.

A recent neural network model of cortical associative memory incorporating neuronal adaptation by a simplified description of its underlying ionic mechanisms is extended towards more realistic network units and architecture. Excitatory units correspond to groups of adapting pyramidal neurons and inhibitory units to groups of nonadapting interneurons. The network architecture is formed from pairs of one pyramidal and one interneuron unit each with inhibitory connections within and excitatory connections between pairs. The degree of adaptability of the pyramidal units controls the character of the network dynamics. An intermediate adaptability generates limit cycles of transitions between stored patterns and regulates oscillation frequencies in the range of theta rhythms observed in the brain. In particular, neuronal adaptation can impose a direction of transitions between overlapping patterns also in a symmetrically connected network. The model permits a detailed analysis of the transition mechanisms. Temporal sequences of patterns thus formed may constitute parts of associative processes, such as recall of stored sequences or search of pattern subspaces. As a special case, neuronal adaptation can accomplish pattern segmentation by which overlapping patterns are temporally resolved. The type of limit cycles produced by neuronal adaptation may also be of significance for central pattern generators, also for networks involving motor neurons. The applied learning rule of Hebbian type is compared to a modified version also common in neural network modelling. It is also shown that the dependence of the network dynamic behaviour on neuronal adaptability, from fixed point attractors at weak adaptability towards more complex dynamics of limit cycles and chaos at strong adaptability, agrees with that recently observed in a more abstract version of the model. The present description of neuronal adaptation is compared to models based on dynamic firing thresholds.

Association↗
PubMedDate not supplied

Application of three-dimensional orthogonal neural network to craniomaxillary reconstruction.

The purpose of this investigation is to establish a practical method to predict and create surface a profile of bone defects by a well-trained 3-D orthogonal neural network. First, the coordinates of the skeletal positions around the boundary of bone defects are input into the 3-D orthogonal neural network to train it to learn the scattering characteristic. The 3-D orthogonal neural network avoids local minima and converges rapidly. After the neural network has been well trained, the mathematic model of the bone defect surface is generated, and the pixel positions are derived. Herein, to verify its performance the proposed method is applied on a patient with a craniofacial defect.

Humans↗
PubMed1996

Neural network differentiation of optic neuritis and anterior ischaemic optic neuropathy.

AIMS: The efficacy of an artificial intelligence technique, neural network analysis, was examined in differentiating two optic neuropathies with overlapping clinical profiles-idiopathic optic neuritis (ON) and non-arteritic anterior ischaemic optic neuropathy (AION). METHODS: A neural network was trained with data from 116 patients with 'gold standard' diagnoses of ON or AION. It was then tested with data from 128 patients with presumed ON or AION, and the correlation of the network's diagnosis with that of expert clinicians tabulated. RESULTS: The network agreed with the clinicians on 97.8% (88 of 90) of the patients with presumed ON and 94.7% (36 of 38) of the patients with presumed AION. Youth, female sex, better initial acuity, a central scotoma, subsequent improvement in acuity, or progressive disease biased the network towards a diagnosis of ON, while advanced age, male sex, presence of hypertension, poor initial acuity, an altitudinal field defect, disc oedema, or less improvement in acuity biased the network towards a diagnosis of AION. CONCLUSION: Neural network analysis is a useful technique for classification of optic neuropathies, particularly where there is overlap of clinical findings.

Adult↗
PubMed1993

Overtraining in neural networks that interpret clinical data.

Backpropagation neural networks are a computer-based pattern-recognition method that has been applied to the interpretation of clinical data. Unlike rule-based pattern recognition, backpropagation networks learn by being repetitively trained with examples of the patterns to be differentiated. We describe and analyze the phenomenon of overtraining in backpropagation networks. Overtraining refers to the reduction in generalization ability that can occur as networks are trained. The clinical application we used was the differentiation of giant cell arteritis (GCA) from other forms of vasculitis (OTH) based on results for 807 patients (593 OTH, 214 GCA) and eight clinical predictor variables. The 807 cases were randomly assigned to either a training set with 404 cases or to a cross-validation set with the remaining 403 cases. The cross-validation set was used to monitor generalization during training. Results were obtained for eight networks, each derived from a different random assignment of the 807 cases. Training error monotonically decreased during training. In contrast, the cross-validation error usually reached a minimum early in training while the training error was still decreasing. Training beyond the minimum cross-validation error was associated with an increased cross-validation error. The shape of the cross-validation error curve and the point during training corresponding to the minimum cross-validation error varied with the composition of the data sets and the training conditions. The study indicates that training error is not a reliable indicator of a network's ability to generalize. To find the point during training when a network generalizes best, one must monitor cross-validation error separately.

Bias↗
PubMed2006

An artificial neural network improves prediction of observed survival in patients with laryngeal squamous carcinoma.

The accepted method of modelling and predicting failure/survival, Cox's proportional hazards model, is theoretically inferior to neural network derived models for analysing highly complex systems with large datasets. A blinded comparison of the neural network versus the Cox's model in predicting survival utilising data from 873 treated patients with laryngeal cancer. These were divided randomly and equally into a training set and a study set and Cox's and neural network models applied in turn. Data were then divided into seven sets of binary covariates and the analysis repeated. Overall survival was not significantly different on Kaplan-Meier plot, or with either test model. Although the network produced qualitatively similar results to Cox's model it was significantly more sensitive to differences in survival curves for age and N stage. We propose that neural networks are capable of prediction in systems involving complex interactions between variables and non-linearity.

Carcinoma, Squamous Cell↗
PubMed2005

Deterministic convergence of an online gradient method for BP neural networks.

Online gradient methods are widely used for training feedforward neural networks. We prove in this paper a convergence theorem for an online gradient method with variable step size for backward propagation (BP) neural networks with a hidden layer. Unlike most of the convergence results that are of probabilistic and nonmonotone nature, the convergence result that we establish here has a deterministic and monotone nature.

Algorithms↗
PubMed2001

Neural network model of gene expression.

Many natural processes consist of networks of interacting elements that, over time, affect each other's state. Their dynamics depend on the pattern of connections and the updating rules for each element. Genomic regulatory networks are networks of this sort. In this paper we use artificial neural networks as a model of the dynamics of gene expression. The significance of the regulatory effect of one gene product on the expression of other genes of the system is defined by a weight matrix. The model considers multigenic regulation including positive and/or negative feedback. The process of gene expression is described by a single network and by two linked networks where transcription and translation are modeled independently. Each of these processes is described by different network controlled by different weight matrices. Methods for computing the parameters of the model from experimental data are discussed. Results computed by means of the model are compared with experimental observations. Generalization to a 'black box' concept, where the molecular processes occurring in the cell are considered as signal processing units forming a global regulatory network, is discussed.-Vohradský, J. Neural network model of gene expression.

Gene Expression Regulation↗
PubMed2006

Multiperiodicity of discrete-time delayed neural networks evoked by periodic external inputs.

In this paper, the multiperiodicity of a general class of discrete-time delayed neural networks (DTDNNs) is formulated and studied. Several sufficient conditions are obtained to ensure n-neuron DTDNNs can have 2n periodic orbits and these periodic orbits are locally attractive. In addition, we give the conditions for a periodic orbit to be locally or globally attractive when the periodic orbit locates in a designated region. As two typical representatives, the Hopfield neural network and the cellular neural network are examined in detail. These conditions improve and extend the existing stability results in the literature. Simulations results are also discussed in three illustrative examples.

Algorithms↗
PubMed2004

Neural networks for protein classification.

This paper describes a biomolecular classification methodology based on multilayer perceptron neural networks. The system developed is used to classify enzymes found in the Protein Data Bank. The primary goal of classification, here, is to infer the function of an (unknown) enzyme by analysing its structural similarity to a given family of enzymes. A new codification scheme was devised to convert the primary structure of enzymes into a real-valued vector. The system was tested with a different number of neural networks, training set sizes and training epochs. For all experiments, the proposed system achieved a higher accuracy rate when compared with profile hidden Markov models. Results demonstrated the robustness of this approach and the possibility of implementing fast and efficient biomolecular classification using neural networks.

Algorithms↗
PubMed1999

Neural network pattern recognition analysis of graft flow characteristics improves intra-operative anastomotic error detection in minimally invasive CABG.

OBJECTIVE: The intra-operative assessment of the quality of anastomosis in minimally invasive coronary artery bypass surgery (CABG) is critical. Recent investigations demonstrated that flow probes used intra-operatively to assess anastomotic errors may give the surgeon a false sense of confidence as only severely stenotic anastomoses (>90%) could be reliably detected. We developed a neural network system using graft flow data and assessed its potential to improve anastomotic error detection. METHODS: Mammary to LAD grafts (n = 46) were constructed in mongrel dogs off-pump. Continuous beat-to-beat graft flow was recorded using transit-time flow probes. Various degrees of anastomotic stenoses (0-100%) were created by an additional suture. The degree of anastomotic stenosis was confirmed by postoperative angiography. A learning vector quantization neural network was created using heart rate, mean aortic pressure, mean systolic, maximum systolic, minimum systolic, mean diastolic, maximum diastolic, minimum diastolic, and mean graft flows. In addition, a spectral analysis of the flow waveforms was performed and the magnitude and phase of the first five harmonics were used to further develop the neural network. RESULTS: The neural network pattern recognition system was 94% accurate in detecting any stenosis >50%. To validate the model, a testing set was used with 20% of the data values, and the accuracy remained at 100% above chance alone. CONCLUSION: Pattern recognition of transit-time flow probe tracings using neural network systems can detect anastomotic errors significantly better than the surgeon's visual assessment, thereby improving the clinical outcome of minimally invasive CABG.

Anastomosis, Surgical↗
PubMed1995

Synapse models for neural networks: from ion channel kinetics to multiplicative coefficient wij.

This paper relates different levels at which the modeling of synaptic transmission can be grounded in neural networks: the level of ion channel kinetics, the level of synaptic conductance dynamics, and the level of a scalar synaptic coefficient. The important assumptions to reduce a synapse model from one level to the next are explicitly exhibited. This coherent progression provides control on what is discarded and what is retained in the modeling process, and is useful to appreciate the significance and limitations of the resulting neural networks. This methodic simplification terminates with a scalar synaptic efficacy as it is very often used in neural networks, but here its conditions of validity are explicitly displayed. This scalar synapse also comes with an expression that directly relates it to basic quantities of synaptic functioning, and it can be endowed with meaningful physical units and realistic numerical values. In addition, it is shown that the scalar synapse does not receive the same expression in neural networks operating with spikes or with firing rates. These coherent modeling elements can help to improve, adjust, and refine the investigation of neural systems and their remarkable collective properties for information processing.

Action Potentials↗
PubMed2000

[Neural network based on modified simplex method and its application in studying forest self-thinning].

The mechanism of forest self-thinning is generally nonlinear and dynamic, and the artificial neural network has the characteristic of expressing arbitrary nonlinear mapping. In this paper, the feasibility and limitation of artificial neural network used to simulating forest self-thinning was expounded, and the principle and algorithms of the neural network model based on modified simplex method (BP-MSM mixed algorithms) for modeling forest self-thinning were described. Its applications in self-thinning of Populus tremula natural forest Cunninghamia lanceolata plantation were illustrated. The results of forest self-thinning examples show the BP-MSM mixed algorithms were satisfactory in simulating forest self-thinning, and its precision was higher, which develops the method and theory of artificial neural network, and enriches the simulating method of forest self-thinning.

Algorithms↗
PubMedDate not supplied

A neural network explanation of posttraumatic stress disorder.

An introduction to connectionistic neural networks and their importance for behavior therapists was provided by Tryon (1995a). This article shows that the bidirectional associative memory connectionistic neural networks model meets all four of Jones and Barlow's (1990) requirements and all five of Brewin, Dalgleish, and Joseph's (1996) requirements for a theoretical explanation of posttraumatic stress disorder (PTSD). The bidirectional associative memory model provides a theoretical basis for explaining the clinical symptom constellation of PTSD, with special emphasis on why trauma is re-experienced through memory; accounts for individual variation in symptom severity, including why some people do not develop PTSD; explains why persons with PTSD are comorbid with depression, generalized anxiety, and substance abuse; explains how abnormal behavior derives from normal memory formation processes; and makes novel predictions.

Association Learning↗