Search PubMed⌕ Search

SEARCH · Search PubMed

Results for “Networks”

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 739 records · Page 41Linked to original sources
PubMedDate not supplied

Stability analysis of higher-order neural networks for combinatorial optimization.

Recurrent neural networks with higher order connections, from here on referred to as higher-order neural networks (HONNs), may be used for the solution of combinatorial optimization problems. In Ref. 5 a mapping of the traveling salesman problem (TSP) onto a HONN of arbitrary order was developed, thereby creating a family of related networks that can be used to solve the TSP. In this paper, we explore the trade-off between network complexity and quality of solution that is made available by the HONN mapping of the TSP. The trade-off is investigated by undertaking an analysis of the stability of valid solutions to the TSP in a HONN of arbitrary order. The techniques used to perform the stability analysis are not new, but have been widely used elsewhere in the literature. The original contribution in this paper is the application of these techniques to a HONN of arbitrary order used to solve the TSP. The results of the stability analysis show that the quality of solution is improved by increasing the network complexity, as measured by the order of the network. Furthermore, it is shown that the Hopfield network, as the simplest network in the family of higher-order networks, is expected to produce the poorest quality of solution.

Algorithms↗
PubMed2004

Generalization of features in the assembly neural networks.

The purpose of the paper is an experimental study of the formation of class descriptions, taking place during learning, in assembly neural networks. The assembly neural network is artificially partitioned into several sub-networks according to the number of classes that the network has to recognize. The features extracted from input data are represented in neural column structures of the sub-networks. Hebbian neural assemblies are formed in the column structure of the sub-networks by weight adaptation. A specific class description is formed in each sub-network of the assembly neural network due to intersections between the neural assemblies. The process of formation of class descriptions in the sub-networks is interpreted as feature generalization. A set of special experiments is performed to study this process, on a task of character recognition using the MNIST database.

Algorithms↗
PubMed1995

Solitary pulmonary nodules: determining the likelihood of malignancy with neural network analysis.

PURPOSE: To test a neural network in differentiation of benign from malignant solitary pulmonary nodules. MATERIALS AND METHODS: Neural networks were trained and tested on the characteristics of 318 nodules. Predictive accuracy of the network was judged for calibration and discrimination. Network results were compared with those with a simpler Bayesian method. RESULTS: The Brier score was 0.142 (calibration, 0.003; discrimination, 0.139) for the neural network and 0.133 for the Bayesian analysis (calibration, 0.012; discrimination, 0.121). Analysis of the calibration curve revealed no significant difference (P < .05) between the slope (b = 1.09) and the line of identity (b = 1) for the neural network or the Bayesian analysis. The area under the receiver operating characteristic curve was 0.871 for the neural network and 0.894 for the Bayesian analysis (P < .05). There were 23 and 21 false-positive predictions and 18 and six false-negative predictions for the neural network and Bayesian analysis, respectively. CONCLUSION: The Bayesian method was better than the neural network in prediction of probability of malignancy in solitary pulmonary nodules.

Adult↗
PubMed1995

Pruning of rat cortical taste neurons by an artificial neural network model.

1. Taste qualities are believed to be coded in the activity of ensembles of taste neurons. However, it is not clear whether all neurons are equally responsible for coding. To clarify the point, the relative contribution of each taste neuron to coding needs to be assessed. 2. We constructed simple three-layer neural networks with input units representing cortical taste neurons of the rat. The networks were trained by the back-propagation learning algorithm to classify the neural response patterns to the basic taste stimuli (sucrose, HCl, quinine hydrochloride, and NaCl). The networks had four output units representing the basic taste qualities, the values of which provide a measure for similarity of test stimuli (salts, tartaric acid, and umami substances) to the basic taste stimuli. 3. Trained networks discriminated the response patterns to the test stimuli in a plausible manner in light of previous physiological and psychological experiments. Profiles of output values of the networks paralleled those of across-neuron correlations with respect to the highest or second-highest values in the profiles. 4. We evaluated relative contributions of input units to the taste discrimination of the network by examining their significance Sj, which is defined as the sum of the absolute values of the connection weights from the jth input unit to the hidden layer. When the input units with weaker connection weights (e.g., 15 of 39 input units) were "pruned" from the trained network, the ability of the network to discriminate the basic taste qualities as well as other test stimuli was not greatly affected. On the other hand, the taste discrimination of the network progressively deteriorated much more rapidly with pruning of input units with stronger connection weights. 5. These results suggest that cortical taste neurons differentially contribute to the coding of taste qualities. The pruning technique may enable the evaluation of a given taste neuron in terms of its relative contribution to the coding, with Sj providing a quantitative measure for such evaluation.

Animals↗
PubMed2000

Attractors: architects of network organization?

An attractor is defined here informally as a state of activity toward which a system settles. The settling or relaxation process dissipates the effects produced by external perturbations. In neural systems the relaxation process occurs temporally in the responses of each neuron and spatially across the network such that the activity settles into a subset of the available connections. Within limits, the set of neurons toward which the coordinated neural firing settles can be different from one time to another, and a given set of neurons can generate different types of attractor activity, depending on how the input environment activates the network. Findings such as these indicate that though information resides in the details of neuroanatomic structure, the expression of this information is in the dynamics of attractors. As such, attractors are sources of information that can be used not only in adaptive behavior, but also to effect the neural architecture that generates the attractor. The discussion here focuses on the latter possibility. A conjecture is offered to show that the relaxation dynamic of an attractor may 'guide' activity-dependent learning processes in such a way that synaptic strengths, firing thresholds, the physical connections between neurons, and the size of the network are automatically set in an optimal, interrelated fashion. This inter-relatedness among network parameters would not be expected from more classical, 'switchboard' approaches to neural integration. The ideas are discussed within the context of 'pulse-propagated networks' or equivalently as 'spike-activated networks' in which the specific order in time intervals between action potentials carries important information for cooperative activity to emerge among neurons in a network. Though the proposed ideas are forward-looking, being based on preliminary work in biological and artificial networks, they are testable in biological neural networks reconstructed from identified neurons in cell culture and in simulation models of them.

Animals↗
PubMed1999

Artificial neural networks applied to survival prediction in breast cancer.

In this study, we evaluated the accuracy of a neural network in predicting 5-, 10- and 15-year breast-cancer-specific survival. A series of 951 breast cancer patients was divided into a training set of 651 and a validation set of 300 patients. Eight variables were entered as input to the network: tumor size, axillary nodal status, histological type, mitotic count, nuclear pleomorphism, tubule formation, tumor necrosis and age. The area under the ROC curve (AUC) was used as a measure of accuracy of the prediction models in generating survival estimates for the patients in the independent validation set. The AUC values of the neural network models for 5-, 10- and 15-year breast-cancer-specific survival were 0.909, 0.886 and 0.883, respectively. The corresponding AUC values for logistic regression were 0.897, 0.862 and 0.858. Axillary lymph node status (N0 vs. N+) predicted 5-year survival with a specificity of 71% and a sensitivity of 77%. The sensitivity of the neural network model was 91% at this specificity level. The rate of false predictions at 5 years was 82/300 for nodal status and 40/300 for the neural network. When nodal status was excluded from the neural network model, the rate of false predictions increased only to 49/300 (AUC 0. 877). An artificial neural network is very accurate in the 5-, 10- and 15-year breast-cancer-specific survival prediction. The consistently high accuracy over time and the good predictive performance of a network trained without information on nodal status demonstrate that neural networks can be important tools for cancer survival prediction.

Adult↗
PubMed2004

A machine learning method for extracting symbolic knowledge from recurrent neural networks.

Neural networks do not readily provide an explanation of the knowledge stored in their weights as part of their information processing. Until recently, neural networks were considered to be black boxes, with the knowledge stored in their weights not readily accessible. Since then, research has resulted in a number of algorithms for extracting knowledge in symbolic form from trained neural networks. This article addresses the extraction of knowledge in symbolic form from recurrent neural networks trained to behave like deterministic finite-state automata (DFAs). To date, methods used to extract knowledge from such networks have relied on the hypothesis that networks' states tend to cluster and that clusters of network states correspond to DFA states. The computational complexity of such a cluster analysis has led to heuristics that either limit the number of clusters that may form during training or limit the exploration of the space of hidden recurrent state neurons. These limitations, while necessary, may lead to decreased fidelity, in which the extracted knowledge may not model the true behavior of a trained network, perhaps not even for the training set. The method proposed here uses a polynomial time, symbolic learning algorithm to infer DFAs solely from the observation of a trained network's input-output behavior. Thus, this method has the potential to increase the fidelity of the extracted knowledge.

Algorithms↗
PubMed2005

Rate and synchrony in feedforward networks of coincidence detectors: analytical solution.

We provide an analytical recurrent solution for the firing rates and cross-correlations of feedforward networks with arbitrary connectivity, excitatory or inhibitory, in response to steady-state spiking input to all neurons in the first network layer. Connections can go between any two layers as long as no loops are produced. Mean firing rates and pairwise cross-correlations of all input neurons can be chosen individually. We apply this method to study the propagation of rate and synchrony information through sample networks to address the current debate regarding the efficacy of rate codes versus temporal codes. Our results from applying the network solution to several examples support the following conclusions: (1) differential propagation efficacy of rate and synchrony to higher layers of a feedforward network is dependent on both network and input parameters, and (2) previous modeling and simulation studies exclusively supporting either rate or temporal coding must be reconsidered within the limited range of network and input parameters used. Our exact, analytical solution for feedforward networks of coincidence detectors should prove useful for further elucidating the efficacy and differential roles of rate and temporal codes in terms of different network and input parameter ranges.

Action Potentials↗
PubMedDate not supplied

Performance of a neural network trained to make third-molar treatment-planning decisions.

The authors developed and tested 12 neural networks of different architectures to make lower-third-molar treatment-planning decisions, using a software-based neural network (Neudesk 1.2, Neural Computer Sciences, Southampton, UK). Network training was undertaken using clinical histories from 119 patients (with 238 lower third molars) referred for treatment planning (79 females and 40 males, mean age 25 years) together with output data consisting of actual treatments planned by a senior oral surgeon. Both the input clinical data and the consultant decisions were treated on a tooth-wise basis and were coded to numerical values. Binary data (e.g., present/absent) were coded to 1 and 0, while quantitative data (e.g., age) were scaled to fall between 0 and 1. A network based on the optimal architecture was trained and then interrogated with test data derived from a further 174 patients (119 females and 55 males, mean age 26 years) with 348 lower third molars. Network decisions were dichotomized with a threshold of 0.8. With no knowledge of the network decisions, the senior oral surgeon indicated his preferred treatments. The teeth were then assigned to "gold-standard" categories of indications present or absent based on National Institutes of Health consensus criteria. Against this, the network achieved a sensitivity of 0.78, which was slightly inferior to that of the oral surgeon (0.88), although this difference was not significant, and a specificity of 0.98, compared with 0.99 for the oral surgeon (p = NS). Agreement between the oral surgeon and network decisions was very high (kappa = 0.850). This study demonstrates that it is possible to train a neural network to provide reliable decision support for lower-third-molar treatment planning.

Adult↗
PubMed2003

The structure of lay consultation networks: managing illness in community settings.

OBJECTIVE: We examined the structure of lay consultation networks among elderly people. METHODS: Data were gathered through interviews with 548 elderly adults living in Florida retirement communities and in Cleveland. Respondents identified people they consulted about symptom or disease information, health worries, what the doctor said, and consulting health providers. Network size, composition, geographic dispersion, gender homogeneity, and division of labor were assessed. RESULTS: Eighty percent identified at least one network member (range = 1 to 7 consultants). Networks largely consisted of family members, particularly spouses and women. Older adults talked most frequently with network members about physician visits. Widowed individuals were more likely to rely on children and friends and have networks outside their neighborhoods than married elders. Women's networks included a broader range of relationships than men's networks. DISCUSSION: Results reaffirmed the importance of gender in structuring networks in late life. The low prevalence of friends supports Cartensen's Selectivity Theory.

Aged↗
PubMed2006

Protein ranking by semi-supervised network propagation.

BACKGROUND: Biologists regularly search DNA or protein databases for sequences that share an evolutionary or functional relationship with a given query sequence. Traditional search methods, such as BLAST and PSI-BLAST, focus on detecting statistically significant pairwise sequence alignments and often miss more subtle sequence similarity. Recent work in the machine learning community has shown that exploiting the global structure of the network defined by these pairwise similarities can help detect more remote relationships than a purely local measure. METHODS: We review RankProp, a ranking algorithm that exploits the global network structure of similarity relationships among proteins in a database by performing a diffusion operation on a protein similarity network with weighted edges. The original RankProp algorithm is unsupervised. Here, we describe a semi-supervised version of the algorithm that uses labeled examples. Three possible ways of incorporating label information are considered: (i) as a validation set for model selection, (ii) to learn a new network, by choosing which transfer function to use for a given query, and (iii) to estimate edge weights, which measure the probability of inferring structural similarity. RESULTS: Benchmarked on a human-curated database of protein structures, the original RankProp algorithm provides significant improvement over local network search algorithms such as PSI-BLAST. Furthermore, we show here that labeled data can be used to learn a network without any need for estimating parameters of the transfer function, and that diffusion on this learned network produces better results than the original RankProp algorithm with a fixed network. CONCLUSION: In order to gain maximal information from a network, labeled and unlabeled data should be used to extract both local and global structure.

Algorithms↗
PubMed2025

Computational network biology analysis revealed COVID-19 severity markers: Molecular interplay between HLA-II with CIITA.

COVID-19, severe acute respiratory syndrome coronavirus 2, rapidly spread worldwide. Severe and critical patients are expected to rapidly deteriorate. Although several studies have attempted to uncover the mechanisms underlying COVID-19 severity, most have focused on the perturbations of single genes. However, the complex mechanism of COVID-19 involves numerous perturbed genes in a molecular network rather than a single abnormal gene. Thus, we aimed to identify COVID-19 severity-specific markers in the Japanese population using gene network analysis. In order to reveal the severity-specific molecular interplays, we developed a novel computational network biology strategy that measures dissimilarity between networks based on the comprehensive information of gene network (i.e., expression levels of genes and network structure) by using Kullback-Leibler divergence. Monte Carlo simulations demonstrated the effectiveness of our strategy for differential gene network analysis. We applied this method to publicly available whole blood RNA-seq data from the Japan coronavirus disease 2019 Task Force and identified differentially regulated molecular interplays between 368 severe and 105 non-severe samples. Our analysis suggests the gene network between HLA class II, CIITA, and CD74 as a COVID-19 severity specific molecular marker. Although the association between HLA class II and COVID-19 has been demonstrated, our data analysis revealed that the molecular interplay of HLA class II with its target and/or regulator is a crucial marker for COVID-19 severity. Our findings from computational network biology analysis suggest that suppression and activation of the molecular interplay between HLA class II, CIITA, and CD74 provide crucial clues to uncover the mechanisms of COVID-19 severity.

Humans↗
PubMedDate not supplied

Cellular dynamics of network memory.

One example of "emergence" is the development, as a result of neural ontogeny and living experience, of cortical networks capable of representing and retaining cognitive information. A large body of evidence from neuropsychology, electrophysiology and neuroimaging indicates that so-called working memory and long-term memory share the same neural substrate in the cerebral cortex. That substrate consists in a system of widespread, overlapping and hierarchically organized networks of cortical neurons. In this system, any neuron or group of neurons can be part of many networks, and thus many memories. Working memory is the temporary activation of one such network of long-term memory for the purpose of executing an action in the near future. The activation of the network may be brought about by stimuli that by virtue of prior experience are in some manner associated with the cognitive content of the network, including the response of the organism to those stimuli. The mechanisms by which the network stays activated are presumed to include the recurrent re-entry of impulses through associated neuronal assemblies of the network. Consistent with this notion is the following evidence: (1) working memory depends on the functional integrity of cortico-cortical connective loops; and (2) during working memory, remarkable similarities--including "attractor behavior"--have been observed between firing patterns in real cortex and in an artificial recurrent network.

Animals↗
PubMed1999

Effect of an artificial neural network on radiologists' performance in the differential diagnosis of interstitial lung disease using chest radiographs.

OBJECTIVE: We developed a new method to distinguish between various interstitial lung diseases that uses an artificial neural network. This network is based on features extracted from chest radiographs and clinical parameters. The aim of our study was to evaluate the effect of the output from the artificial neural network on radiologists' diagnostic accuracy. MATERIALS AND METHODS: The artificial neural network was designed to differentiate among 11 interstitial lung diseases using 10 clinical parameters and 16 radiologic findings. Thirty-three clinical cases (three cases for each lung disease) were selected. In the observer test, chest radiographs were viewed by eight radiologists (four attending physicians and four residents) with and without network output, which indicated the likelihood of each of the 11 possible diagnoses in each case. The radiologists' performance in distinguishing among the 11 interstitial lung diseases was evaluated by receiver operating characteristic (ROC) analysis with a continuous rating scale. RESULTS: When chest radiographs were viewed in conjunction with network output, a statistically significant improvement in diagnostic accuracy was achieved (p < .0001). The average area under the ROC curve was .826 without network output and .911 with network output. CONCLUSION: An artificial neural network can provide a useful "second opinion" to assist radiologists in the differential diagnosis of interstitial lung disease using chest radiographs.

Diagnosis, Differential↗
PubMed2000

How well can radiologists using neural network software diagnose pulmonary embolism?

OBJECTIVE: This study evaluated and optimized the performance of an automated artificial neural network image interpreter in the diagnosis of pulmonary embolism on ventilation-perfusion lung scans. The computer interpretations were compared with the interpretations of three experienced observers. MATERIALS AND METHODS: Digital data were obtained from 100 patients with normal findings on chest radiographs who were undergoing both radionuclide ventilation-perfusion scanning and pulmonary angiography. Interpretations of differently trained neural networks were compared with those of three experienced nuclear medicine practitioners unaware of the clinical diagnosis. RESULTS: Machines running neural networks performed similarly to experienced scan interpreters in the detection of pulmonary embolism. Both the human observers and the networks performed best in cases with large emboli. Neural network performance was best in the right lung, when the networks were trained using only cases with large emboli and when networks were trained independently in the right and left lungs. The best predictions resulted from a collaborative interpretation incorporating both the human and computer predictions. CONCLUSION: Computers running artificial neural networks using scan data obtained directly from the anterior and posterior ventilation and perfusion images, without human involvement, perform comparably with experienced observers in patients with normal findings on chest radiographs. Human observers can improve their interpretations by incorporating computer output to formulate diagnostic prediction. The method of training the networks is critical to optimizing performance.

Adult↗
PubMed2000

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

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

Absorptiometry, Photon↗
PubMed1997

Neural networks in neurotologic expert systems.

Artificial intelligence donates new possibilities to neurotologic research. Neural networks are a computer-based reasoning method which can be applied in expert systems created for clinical decision support. Neural networks have been used in medical imaging, in medical signal processing and to analyze both clinical and laboratory data. Principally, neural networks simulate the function of the brain. They have to be taught to make correct decisions from the input data. This learning process can be either supervised or unsupervised. The decision making is based on mathematical transformations and it occurs on a hidden level. Calculations are made on parallel manner and the decision making simulates pattern recognition method. Neural networks suit well in medical problems which cannot be defined in simple rules. A drawback of neural networks is that the decisions are irrational and cannot be motivated to the user. Another problem is neural networks' difficulty to handle incomplete input data, i.e., how to define some default or expected values for unknown input parameters. In a complex medical area, which would require multilayered neural networks, the neural networks require a large amount of solved cases for the learning process. In our experience neural networks seem not suitable for diagnosing vertigo and a better choice would be either case-based reasoning or possibly genetic algorithms or a combination of these.

Diagnosis, Computer-Assisted↗
PubMedDate not supplied

A personal computer network system for equitable allocation of cadaver organs.

We developed a personal computer network system for the equitable allocation of cadaveric organs. This network consists of a host computer (IBM PS55 model 5570 T) and various kinds of personal computers manufactured by many different computer makers in Japan. The merits of our personal computer network include lower cost and an easy access to the host computer from all the centres participating in this network while using their own favourite personal computers. Among the programs made for allocating cadaveric organs, we present in this paper the program for livers. This program was developed with a modified version of the logic developed by Starzl et al. The grade modification for the United Network for Organ Sharing (UNOS) in the United States was used as the basis for classification of medical urgency. Our program weighed the factors of medical urgency, compatibility of blood group and waiting time. Distance factors were omitted because of the smaller area of the network compared to that of UNOS. This computer network would be linked to other computer networks in creating a national organ procurement and transplant network in Japan, in order to help them to catch up with other advanced transplant countries. Such an equal and objective computer system should allow organ transplantation to become more widely accepted.

Computer Communication Networks↗