PubMed Health⌕ Search

SEARCH · PubMed Health

Results for “Neural Networks, Computer”

Explore indexed PubMed citations for clinical trials, systematic reviews and public health research. Read source abstracts and follow each citation to its original PubMed record.

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

At least 595 records · Page 33Linked to original sources
PubMed1995

Computer-aided detection of mammographic microcalcifications: pattern recognition with an artificial neural network.

We are developing a computer program for automated detection of clustered microcalcifications on mammograms. In this study, we investigated the effectiveness of a signal classifier based on a convolution neural network (CNN) approach for improvement of the accuracy of the detection program. Fifty-two mammograms with clustered microcalcifications were selected from patient files. The clusters on the mammograms were ranked by experienced mammographers and divided into an obvious group, an average group, and a subtle group. The average and subtle groups were combined and randomly divided into two sets, each of which was used as training or test set alternately. The obvious group served as an additional independent test set. Regions of interest (ROIs) containing potential individual microcalcifications were first located on each mammogram by the automated detection program. The ROIs from one set of the mammograms were used to train CNNs of different configurations with a back-propagation method. The generalization capability of the trained CNNs was then examined by their accuracy of classifying the ROIs from the other set and from the obvious group. The classification accuracy of the CNNs for the ROIs was evaluated by receiver operating characteristic (ROC) analysis. It was found that CNNs of many different configurations can reach approximately the same performance level, with the area under the ROC curve (Az) of 0.9. We incorporated a trained CNN into the detection program and evaluated the improvement of the detection accuracy by the CNN using free response ROC analysis. Our results indicated that, over a wide range of true-positive (TP) cluster detection rate, the CNN classifier could reduce the number of false-positive (FP) clusters per image by more than 70%. For the obvious cases, at a TP rate of 100%, the FP rate reduced from 0.35 cluster per image to 0.1 cluster per image. For the average and subtle cases, the detection accuracy improved from a TP rate of 87% at an FP rate of four clusters per image to a TP rate of 90% at an FP rate of 1.5 clusters per image.

Automation↗
PubMedDate not supplied

An efficient method for detecting connectivity in neural ensembles.

Modern technology is allowing researchers to collect data from neural ensembles with a large number of units, and the analysis of interaction between these units can be very time consuming. Estimation of pairwise connectivity is the most common method of determining the neural 'network' but usually necessitates the production of numerous histograms for each pair considered. We present a method which will indicate which pairs in a network represent potential connections and thereby simplify the postexperimental analysis. The technique uses cross-interval information to create an n x n matrix which represents all possible connections in an n neuron ensemble and can be calculated recursively on-line. The performance of this technique is analyzed with respect to data size and strength of the connections. It is compared to 2 similar techniques that are also presented here, one in which perfect knowledge of the timing of the excitation is known, and one in which the timing can be bounded.

Computer Simulation↗
PubMed1998

A simple model for neural computation with firing rates and firing correlations.

A simple extension of standard neural network models is introduced which provides a model for neural computations that involve both firing rates and firing correlations. Such an extension appears to be useful since it has been shown that firing correlations play a significant computational role in many biological neural systems. Standard neural network models are only suitable for describing neural computations in terms of firing rates. The resulting extended neural network models are still relatively simple, so that their computational power can be analysed theoretically. We prove rigorous separation results, which show that the use of firing correlations in addition to firing rates can drastically increase the computational power of a neural network. Furthermore, one of our separation results also throws new light on a question that involves just standard neural network models: we prove that the gap between the computational power of high-order and first-order neural nets is substantially larger than shown previously.

Neural Networks, Computer↗
PubMed1993

Artificial neural networks as adjuncts for assessing medical students' problem solving performances on computer-based simulations.

Artificial neural networks were trained by supervised learning to recognize the test selection patterns associated with students' successful solutions to seven immunology computer-based simulations. New test selection patterns evaluated by the trained neural network were correctly classified as successful or unsuccessful solutions to the problem > 90% of the time. The examination of the neural networks output weights after each test selection revealed a progressive and selective increase for the relevant problem suggesting that a successful solution is represented by the neural network as the accumulation of relevant tests. Unsuccessful problem solutions were classified by the neural network software into two patterns of students performance. The first pattern was characterized by low neural network output weights for all seven problems reflecting extensive searching and lack of recognition of relevant information. In the second pattern, the output weights from the neural network were biased toward one of the remaining six incorrect problems suggesting that the student misrepresented the current problem as an instance of a previous problem. Finally, neural network analysis could detect cases where the students switched hypotheses during the problem solving exercises.

California↗
PubMed2001

The portable UNIX programming system (PUPS) and CANTOR: a computational environment for dynamical representation and analysis of complex neurobiological data.

Many problems in analytical biology, such as the classification of organisms, the modelling of macromolecules, or the structural analysis of metabolic or neural networks, involve complex relational data. Here, we describe a software environment, the portable UNIX programming system (PUPS), which has been developed to allow efficient computational representation and analysis of such data. The system can also be used as a general development tool for database and classification applications. As the complexity of analytical biology problems may lead to computation times of several days or weeks even on powerful computer hardware, the PUPS environment gives support for persistent computations by providing mechanisms for dynamic interaction and homeostatic protection of processes. Biological objects and their interrelations are also represented in a homeostatic way in PUPS. Object relationships are maintained and updated by the objects themselves, thus providing a flexible, scalable and current data representation. Based on the PUPS environment, we have developed an optimization package, CANTOR, which can be applied to a wide range of relational data and which has been employed in different analyses of neuroanatomical connectivity. The CANTOR package makes use of the PUPS system features by modifying candidate arrangements of objects within the system's database. This restructuring is carried out via optimization algorithms that are based on user-defined cost functions, thus providing flexible and powerful tools for the structural analysis of the database content. The use of stochastic optimization also enables the CANTOR system to deal effectively with incomplete and inconsistent data. Prototypical forms of PUPS and CANTOR have been coded and used successfully in the analysis of anatomical and functional mammalian brain connectivity, involving complex and inconsistent experimental data. In addition, PUPS has been used for solving multivariate engineering optimization problems and to implement the digital identification system (DAISY), a system for the automated classification of biological objects. PUPS is implemented in ANSI-C under the POSIX.1 standard and is to a great extent architecture- and operating-system independent. The software is supported by systems libraries that allow multi-threading (the concurrent processing of several database operations), as well as the distribution of the dynamic data objects and library operations over clusters of computers. These attributes make the system easily scalable, and in principle allow the representation and analysis of arbitrarily large sets of relational data. PUPS and CANTOR are freely distributed (http://www.pups.org.uk) as open-source software under the GNU license agreement.

Algorithms↗
PubMed2025

HallmarkGraph: a cancer hallmark informed graph neural network for classifying hierarchical tumor subtypes.

MOTIVATION: Accurate tumor subtype diagnosis is crucial for precision oncology, yet current methodologies face significant challenges. These include balancing model accuracy with interpretability and the high costs of generating multi-omics data in clinical settings. Moreover, there is a lack of validated models capable of classifying hierarchical tumor subtypes across a comprehensive pan-cancer cohort. RESULTS: We present a graph neural network, HallmarkGraph, the first biologically informed model developed to classify hierarchical tumor subtypes in human cancer. Inspired by cancer hallmarks, the model's architecture integrates transcriptome profiles and gene regulatory interactions to perform multi-label classification. We evaluate the model on a comprehensive pan-cancer cohort comprising 11 476 samples from 26 primary cancers with 405 subtypes up to eight levels. The model demonstrates exceptional performance, achieving 5-fold cross-validation accuracy between 85% and 99% for tumor subtypes labeled with increasing details of genomic information. It also shows good generalizability on a validation dataset of 887 samples, assessed using three metrics that consider tumor subtypes at individual, combined, and sample levels. Benchmarking and ablation experiments show that hallmark-based embeddings slightly influence model performance, while the integrated multilayer perceptron plays a significant role in determining classifier accuracy. Additionally, we use the SHAP method to link cancer hallmarks with genes, identifying key features that influence model decisions. Our findings present a biologically informed machine learning framework capable of tracking tumor transcriptomic trajectories and distinguishing inter- and intra-tumor heterogeneity in pan-cancer. This approach holds promise for enhancing cancer diagnostics. AVAILABILITY AND IMPLEMENTATION: HallmarkGraph is accessible at https://github.com/laixn/HallmarkGraph.

Humans↗
PubMed1993

A genetic algorithm to improve a neural network to predict a patient's response to warfarin.

The ability of neural networks to predict the international normalised ratio (INR) for patients treated with Warfarin was investigated. Neural networks were obtained by using all the predictor variables in the neural network, or by using a genetic algorithm to select an optimal subset of predictor variables in a neural network. The use of a genetic algorithm gave a marked and significant improvement in the prediction of the INR in two of the three cases investigated. The mean error in these cases, typically, reduced from 1.02 +/- 0.29 to 0.28 +/- 0.25 (paired t-test, t = -4.71, p < 0.001, n = 30). The use of a genetic algorithm with Warfarin data offers a significant enhancement of the predictive ability of a neural network with Warfarin data, identifies significant predictor variables, reduces the size of the neural network and thus the speed at which the reduced network can be trained, and reduces the sensitivity of a network to over-training.

Algorithms↗
PubMed1999

Myocardial SPET: artificial neural networks describe extent and severity of perfusion defects.

Artificial neural networks are computer programs that learn from examples. They have been successfully used to detect coronary artery disease from myocardial perfusion images. The purpose of the present study was to develop neural networks that could classify myocardial scintigrams regarding reversibility, localization, severity and extent of perfusion defects. Rest/exercise technetium-99m sestamibi scintigrams from 338 patients were studied. The classifications of two experts were employed as the gold standard. Artificial neural networks were trained to classify both reversible (ischaemia) and non-reversible (infarct) defects in three vascular territories, corresponding to the main coronary arteries. The extent (small or large) and severity (mild or severe) of the defects were described by the networks. After the training process, separate test sets were used to compare the neural networks with one of the experts who reclassified the scintigrams two months later. The neural networks made correct classifications in 71% of the test cases and the human expert in 70% (P=0.10). It was concluded that artificial neural networks can be trained to make clinical interpretations of myocardial perfusion scintigrams. The results indicate that networks can assist physicians in achieving correct interpretations and thereby improve the diagnostic accuracy of medical imaging.

Adult↗
PubMed1996

What can artificial neural networks teach us about neurodegenerative disorders with extrapyramidal features?

Artificial neural networks (ANNs), computer paradigms that can learn, excel in pattern recognition tasks such as disease diagnosis. Artificial neural networks operate in two different learning modes: supervised, in which a known diagnostic outcome is presented to the ANN, and unsupervised, in which the diagnostic outcome is not presented. A supervised learning ANN could emulate human expert diagnostic performance and identify relevant predictive markers in the diagnostic task, while an unsupervised learning ANN could suggest reasonable alternative diagnostic classification criteria. In the present study, we used ANN methodology to try to overcome the neuropathological difficulties in differentiating the subtypes of progressive supranuclear palsy (PSP), and in differentiating PSP from postencephalitic parkinsonism (PEP) and corticobasal degeneration, or Pick's disease from corticobasal degeneration. First, we applied supervised learning ANN to classify 62 cases of these disorders and to identify diagnostic markers that distinguish them. In a second experiment, we used unsupervised learning ANN to investigate possible alternative nosological classifications. Artificial neural networks input data for each case consisted of values representing histological features, including neurofibrillary tangles, neuronal loss and gliosis found in multiple brain sampling areas. The supervised learning ANN achieved excellent accuracy in classifying PSP but had difficulty classifying the other disorders. This method identified a few features that might help to differentiate PEP, supported currently proposed criteria for Pick's disease, corticobasal degeneration and typical PSP, but detected no features to characterize the atypical subtype of PSP. In general, unsupervised learning ANN supported the present nosological classification for PSP, PEP, Pick's disease and corticobasal degeneration, although it overlapped some groups. Artificial neural networks methodology appears promising for studying neurodegenerative disorders.

Basal Ganglia Diseases↗
PubMed1996

Prospective validation of artificial neural network trained to identify acute myocardial infarction.

BACKGROUND: Artificial neural networks apply non-linear statistics to pattern recognition problems. One such problem is acute myocardial infarction (AMI), a diagnosis which, in a patient presenting as an emergency, can be difficult to confirm. We report here a prospective comparison of the diagnostic accuracy of a network and that of physicians, on the same patients with suspected AMI. METHODS: Emergency department physicians who evaluated 1070 patients 18 years or older presenting to the emergency department of a teaching hospital in California, USA with anterior chest pain indicated whether they thought these patients had sustained a myocardial infarction. The network analysed the patient data collected by the physicians during their evaluations and also generated a diagnosis. FINDINGS: The physicians had a diagnostic sensitivity and specificity for myocardial infarction of 73.3% (95% confidence interval 63.3-83.3%) and 81.1% (78.7-83.5%), respectively, while the network had a diagnostic sensitivity and specificity of 96.0% (91.2-100%) and 96.0% (94.8-97.2%), respectively. Only 7% of patients had had an AMI, a low frequency but typical for anterior chest pain. INTERPRETATION: The application of non-linear neural computational analysis via an artificial neural network to the clinical diagnosis of myocardial infarction appears to have significant potential.

Chest Pain↗
PubMed2005

Hierarchical Kohonenen net for anomaly detection in network security.

A novel multilevel hierarchical Kohonen Net (K-Map) for an intrusion detection system is presented. Each level of the hierarchical map is modeled as a simple winner-take-all K-Map. One significant advantage of this multilevel hierarchical K-Map is its computational efficiency. Unlike other statistical anomaly detection methods such as nearest neighbor approach, K-means clustering or probabilistic analysis that employ distance computation in the feature space to identify the outliers, our approach does not involve costly point-to-point computation in organizing the data into clusters. Another advantage is the reduced network size. We use the classification capability of the K-Map on selected dimensions of data set in detecting anomalies. Randomly selected subsets that contain both attacks and normal records from the KDD Cup 1999 benchmark data are used to train the hierarchical net. We use a confidence measure to label the clusters. Then we use the test set from the same KDD Cup 1999 benchmark to test the hierarchical net. We show that a hierarchical K-Map in which each layer operates on a small subset of the feature space is superior to a single-layer K-Map operating on the whole feature space in detecting a variety of attacks in terms of detection rate as well as false positive rate.

Algorithms↗
PubMed2006

A fast identification algorithm for Box-Cox transformation based radial basis function neural network.

In this letter, a Box-Cox transformation-based radial basis function (RBF) neural network is introduced using the RBF neural network to represent the transformed system output. Initially a fixed and moderate sized RBF model base is derived based on a rank revealing orthogonal matrix triangularization (QR decomposition). Then a new fast identification algorithm is introduced using Gauss-Newton algorithm to derive the required Box-Cox transformation, based on a maximum likelihood estimator. The main contribution of this letter is to explore the special structure of the proposed RBF neural network for computational efficiency by utilizing the inverse of matrix block decomposition lemma. Finally, the Box-Cox transformation-based RBF neural network, with good generalization and sparsity, is identified based on the derived optimal Box-Cox transformation and a D-optimality-based orthogonal forward regression algorithm. The proposed algorithm and its efficacy are demonstrated with an illustrative example in comparison with support vector machine regression.

Algorithms↗
PubMed2004

Computer control of pH and DO in a laboratory fermenter using a neural network technique.

In this contribution, the advantages of the artificial neural network approach to the identification and control of a laboratory-scale biochemical reactor are demonstrated. It is very important to be able to maintain the levels of two process variables, pH and dissolved oxygen (DO) concentration, over the course of fermentation in biosystems control. A PC-supported, fully automated, multi-task control system has been designed and built by the authors. Forward and inverse neural process models are used to identify and control both the pH and the DO concentration in a fermenter containing a Saccharomyces cerevisiae based-culture. The models are trained off-line, using a modified back-propagation algorithm based on conjugate gradients. The inverse neural controller is augmented by a new adaptive term that results in a system with robust performance. Experimental results have confirmed that the regulatory and tracking performances of the control system proposed are good.

Algorithms↗
PubMed1992

Predicting protein secondary structure content. A tandem neural network approach.

A priori knowledge of secondary structure content can be of great use in theoretical and experimental determination of protein structure. We present a method that uses two computer-simulated neural networks placed in "tandem" to predict the secondary structure content of water-soluble, globular proteins. The first of the two networks, NET1, predicts a protein's helix and strand content given information about the protein's amino acid composition, molecular weight and heme presence. Because NET1 contained more adjustable parameters (network weights) than learning examples, this network experienced problems with memorization, which is the inability to generalize onto new, never-seen-before examples. To overcome this problem, we designed a second network, NET2, which learned to determine when NET1 was in a state of generalization. Together, these two networks produce prediction errors as low as 5.0% and 5.6% for helix and strand content, respectively, on a set of protein crystal structures bearing little homology to those used in network training. A comparison between three other methods including a multiple linear regression analysis, a non-hidden-node network analysis and a secondary structure assignment analysis reveals that our tandem neural network scheme is, indeed, the best method for predicting secondary structure content. The results of our analysis suggest that the knowledge of sequence information is not necessary for highly accurate predictions of protein secondary structure content.

Databases, Bibliographic↗
PubMed2001

Estimation of spinal deformity in scoliosis from torso surface cross sections.

STUDY DESIGN: Correlation of torso scan and three-dimensional radiographic data in 65 scans of 40 subjects. OBJECTIVES: To assess whether full-torso surface laser scan images can be effectively used to estimate spinal deformity with the aid of an artificial neural network. SUMMARY OF BACKGROUND DATA: Quantification of torso surface asymmetry may aid diagnosis and monitoring of scoliosis and thereby minimize the use of radiographs. Artificial neural networks are computing tools designed to relate input and output data when the form of the relation is unknown. METHODS: A three-dimensional torso scan taken concurrently with a pair of radiographs was used to generate an integrated three-dimensional model of the spine and torso surface. Sixty-five scan-radiograph pairs were generated during 18 months in 40 patients (Cobb angles 0-58 degrees ): 34 patients with adolescent idiopathic scoliosis and six with juvenile scoliosis. Sixteen (25%) were randomly selected for testing and the remainder (n = 49) used to train the artificial neural network. Contours were cut through the torso model at each vertebral level, and the line joining the centroids of area of the torso contours was generated. Lateral deviations and angles of curvature of this line, and the relative rotations of the principal axes of each contour were computed. Artificial neural network estimations of maximal computer Cobb angle were made. RESULTS: Torso-spine correlations were generally weak (r < 0.5), although the range of torso rotation related moderately well to the maximal Cobb angle (r = 0.64). Deformity of the torso centroid line was minimal despite significant spinal deformity in the patients studied. Despite these limitations and the small data set, the artificial neural network estimated the maximal Cobb angle within 6 degrees in 63% of the test data set and was able to distinguish a Cobb angle greater than 30 degrees with a sensitivity of 1.0 and specificity of 0.75. CONCLUSIONS: Neural-network analysis of full-torso scan imaging shows promise to accurately estimate scoliotic spinal deformity in a variety of patients.

Adolescent↗
PubMed1997

Prediction of posterior fossa tumor type in children by means of magnetic resonance image properties, spectroscopy, and neural networks.

Recent studies have explored characteristics of brain tumors by means of magnetic resonance spectroscopy (MRS) to increase diagnostic accuracy and improve understanding of tumor biology. In this study, a computer-based neural network was developed to combine MRS data (ratios of N-acetyl-aspartate, choline, and creatine) with 10 characteristics of tumor tissue obtained from magnetic resonance (MR) studies, as well as tumor size and the patient's age and sex, in hopes of further improving diagnostic accuracy. Data were obtained in 33 children presenting with posterior fossa tumors. The cases were analyzed by a neuroradiologist, who then predicted the tumor type from among three categories (primitive neuroectodermal tumor, astrocytoma, or ependymoma/other) based only on the data obtained via MR imaging. These predictions were compared with those made by neural networks that had analyzed different combinations of the data. The neuroradiologist correctly predicted the tumor type in 73% of the cases, whereas four neural networks using different datasets as inputs were 58 to 95% correct. The neural network that used only the three spectroscopy ratios had the least predictive ability. With the addition of data including MR imaging characteristics, age, sex, and tumor size, the network's accuracy improved to 72%, consistent with the predictions of the neuroradiologist who was using the same information. Use of only the analog data (leaving out information obtained from MR imaging), resulted in 88% accuracy. A network that used all of the data was able to identify 95% of the tumors correctly. It is concluded that a neural network provided with imaging data, spectroscopic data, and a limited amount of clinical information can predict pediatric posterior fossa tumor type with remarkable accuracy.

Astrocytoma↗
PubMed1999

Recognition of seed varieties using a temporal organisation map analysis of electrophoretic images.

This paper presents a method for seed varieties recognition using one-dimensional electrophoresis gels. It employs a neural network basically constituted of temporal organisation maps (TOM). The TOM model is a neural net which was initially developed for speech recognition. It can be trained to recognise words in speech by reference to the sound pattern over a sequence of time steps. Electrophoresis creates a set of bands in the gel, caused by migration of protein from the seed. Each seed variety generates a characteristic pattern. The bands are made visible by staining. They can then be imaged and digitised to create an input to a TOM, which treats the variation with distance along the lane in the same way as the time sequence for which it was originally employed. In this way the characteristic signature of a seed variety can be recognised. A set of 50 images--each containing 10 to 15 lanes--was used to train and test the performance of a neural network in recognising 75 cereal varieties. The network could achieve a recognition rate of 98%, provided that the gel was not distorted or cracked during heating or drying. Details of the design and training of the network are given.

Artifacts↗