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 1,189 records · Page 66Linked to original sources
PubMed2000

Visualizing the function computed by a feedforward neural network.

A method for visualizing the function computed by a feedforward neural network is presented. It is most suitable for models with continuous inputs and a small number of outputs, where the output function is reasonably smooth, as in regression and probabilistic classification tasks. The visualization makes readily apparent the effects of each input and the way in which the functions deviate from a linear function. The visualization can also assist in identifying interactions in the fitted model. The method uses only the input-output relationship and thus can be applied to any predictive statistical model, including bagged and committee models, which are otherwise difficult to interpret. The visualization method is demonstrated on a neural network model of how the risk of lung cancer is affected by smoking and drinking.

Alcohol Drinking↗
PubMed1997

Evaluation of complications of kidney transplantation using artificial neural networks.

The aim of this study was to develop an artificial neural network (ANN) to differentiate between rejection, acute tubular necrosis (ATN) and normally functioning kidneys in a group of patients with renal transplants. The performance of ANN was compared with that of an experienced observer using a database of 35 patients' records, each of which included 12 quantitative parameters derived from renograms and clinical data as well as a clinical evaluation. These findings were encoded as features for a three-layered neural network to predict the outcome of biopsy or clinical diagnosis. The network was trained and tested using the jackknife method and its performance was then compared to that of a radiologist. The network was able to correctly classify 31 of the 35 original cases and gave a better diagnostic accuracy (88%) than the radiologist (83%), by showing an association between the quantitative data and the corresponding pathological results (r = 0.78, P < 0.001). We conclude that an ANN can be trained to differentiate rejection from acute tubular necrosis, as well as normally functioning transplants, with a reasonable degree of accuracy.

Adult↗
PubMed1995

Neural network control of functional neuromuscular stimulation systems: computer simulation studies.

A neural network control system has been designed for the control of cyclic movements in Functional Neuromuscular Stimulation (FNS) systems. The design directly addresses three major problems in FNS control systems: customization of control system parameters for a particular individual, adaptation during operation to account for changes in the musculoskeletal system, and attaining resistance to mechanical disturbances. The control system was implemented by a two-stage neural network that utilizes a combination of adaptive feedforward and feedback control techniques. A new learning algorithm was developed to provide rapid customization and adaptation. The control system was evaluated in a series of studies on a computer simulated musculoskeletal model. The model of electrically stimulated muscle used in the study included nonlinear recruitment, linear dynamics, and multiplicative nonlinear torque-angle and torque-velocity scaling factors. The skeletal model consisted of a one-segment planar system with passive constraints on joint movement. Results of the evaluation have demonstrated that the control system can provide automated customization of the feedforward controller parameters for a given musculoskeletal system. It can account for changes in the musculoskeletal system by adapting the feedforward controller parameters on-line and it can resist the effects of mechanical disturbances. These results suggest that this design may be suitable for the control of FNS systems and other dynamic systems.

Adaptation, Physiological↗
PubMedDate not supplied

The accuracy of artificial neural networks in predicting long-term outcome after traumatic brain injury.

OBJECTIVE: This study compared the accuracy of artificial neural networks to multiple regression and classification and regression trees in predicting outcomes of 1,644 patients in the Traumatic Brain Injury Model Systems database 1 year after injury. METHODS: Data from rehabilitation admission were used to predict discharge scores on the Functional Independence Measure, the Disability Rating Scale, and the Community Integration Questionnaire. RESULTS: Artificial neural networks did not demonstrate greater accuracy in predicting outcomes than did the more widely used method of multiple regression. Both of these methods outperformed classification and regression trees. CONCLUSION: Because of the sophisticated form of multiple regression with splines that was used, firm conclusions are limited about the relative accuracy of artificial neural networks compared to more widely used forms of multiple regression.

Brain Injuries↗
PubMed1995

GENESIS as a neural network modeling tool for saccadic eye movements.

Interest in exploring the computational properties of networks of parallel distributed processing elements has increased due to studies of the structure and function of biological neural networks. Much of the current research involves numerical simulations of these types of networks. GENESIS (GEneral NEtwork SImulation System) and XODUS (X-based Ouput and Display Utility for Simulators) are a result of that interest. The two systems combined provides a standardized and flexible means of constructing neural networks. The structure of GENESIS follows the actual biological network much more closely than artificial neural networks. This paper presents a simple simulation of the cable properties of a dendritic cable consisting of twenty compartments and one soma. The results of the simulation are then compared to actual cable metrics.

Animals↗
PubMed1995

A feed forward neural network for classification of bull's-eye myocardial perfusion images.

Identification of hypoperfused areas in myocardial perfusion single-photon emission tomography studies can be aided by bull's-eye representation of raw counts, lesion extent and lesion severity, the latter two being produced by comparison of the raw bull's-eye data with a normal data base. An artificial intelligence technique which is presently becoming widely popular and which is particularly suitable for pattern recognition is that of artificial neural network. We have studied the ability of feed forward neural networks to extract patterns from bull's-eye data by assessing their capability to predict lesion presence without direct comparison with a normal data base. Studies were undertaken on both simulation data and on real stress-rest data obtained from 410 male patients undergoing routine thallium-201 myocardial perfusion scintigraphy. The ability of trained neural networks to predict lesion presence was quantified by calculating the areas under receiver operating characteristic curves. Figures as high as 0.96 for non-preclassified patient data were obtained, corresponding to an accuracy of 92%. The results demonstrate that neural networks can accurately classify patterns from bull's-eye myocardial perfusion images and detect the presence of hypoperfused areas without the need for comparison with a normal data base. Preliminary work suggests that this technique could be used to study perfusion patterns in the myocardium and their correlation with clinical parameters.

Computer Simulation↗
PubMed1994

Automated interpretation of planar thallium-201-dipyridamole stress-redistribution scintigrams using artificial neural networks.

UNLABELLED: To develop an automated image interpretation system of planar cardiac 201Tl dipyridamole stress/redistribution scintigrams, the authors used artificial neural networks that associate patterns of segmental myocardial thallium uptake with a diagnostic assessment about the presence, severity and localization of significant coronary artery disease. METHODS: Artificial neural networks were trained and evaluated using the results from segmental thallium analysis and either expert readings in 159 cases or coronary angiography in a subgroup of 81 patients. RESULTS: Based on receiver operating characteristics analysis, the sensitivity for the detection of significant coronary artery disease at a specificity of 90% was 51% compared with angiography and 72% compared with the human expert. For severity and localization of disease, two vascular territories assigned to the vascular bed of the left anterior descending (LAD) artery and to the territory subtended by the left circumflex artery and the right coronary artery together (CX/RCA) were included in the analysis. CONCLUSION: Artificial neural networks may be useful to develop automated computer-based image interpretation systems of 201Tl perfusion scintigrams. However, utilization of large training datasets appears to be a prerequisite to achieve adequate diagnostic performance.

Adult↗
PubMed1993

Applications of neural networks in structure-activity relationships of a small number of molecules.

We investigated the applications of back propagation artificial neural networks (ANN) for a small dataset analysis in the field of structure-activity relationships. The derivatives of carboquinone were used as an example. It's been found that in this case the use of the same neural network results in unambiguous classification of new molecules. Predictions can be improved with statistical analysis of independent prognosis sets. We suggest that the sign criterion be used as a classification rule. We also compared neural networks with FALS and ALS in leave-one-out prediction. ANN applied to the same dataset has shown the same predictive ability as ALS but poorer than FALS.

Carbazilquinone↗
PubMed2004

Simultaneous determination of aniline and cyclohexylamine by principal component artificial neural networks.

A specterophotometric method for simultaneous determination of aniline and cyclohexylamine using principal component artificial neural networks is proposed. This method is based on the reactions involving aniline and/or cyclohexylamine, with bis(acetylacetoneethylendiamine)tributylphosphine cobalt(III) perchlorate as a complexing reagent. A nonionic surfactant, Triton X-100, was used for dissolving the complexes and intensifying the signals. The absorption data were based on the spectra registered in the range of 350 - 550 nm. An artificial neural network consisting of three layers of nodes was trained by applying a back-propagation learning rule. Sigmoid transfer functions were used in the hidden and output layers to facilitate nonlinear calibration. The predictive ability of artificial neural networks was examined for the determination of aniline and cyclohexylamine in synthetic mixtures.

Journal Article↗
PubMedDate not supplied

Comparison of support vector machine and artificial neural network systems for drug/nondrug classification.

Support vector machine (SVM) and artificial neural network (ANN) systems were applied to a drug/nondrug classification problem as an example of binary decision problems in early-phase virtual compound filtering and screening. The results indicate that solutions obtained by SVM training seem to be more robust with a smaller standard error compared to ANN training. Generally, the SVM classifier yielded slightly higher prediction accuracy than ANN, irrespective of the type of descriptors used for molecule encoding, the size of the training data sets, and the algorithm employed for neural network training. The performance was compared using various different descriptor sets and descriptor combinations based on the 120 standard Ghose-Crippen fragment descriptors, a wide range of 180 different properties and physicochemical descriptors from the Molecular Operating Environment (MOE) package, and 225 topological pharmacophore (CATS) descriptors. For the complete set of 525 descriptors cross-validated classification by SVM yielded 82% correct predictions (Matthews cc = 0.63), whereas ANN reached 80% correct predictions (Matthews cc = 0.58). Although SVM outperformed the ANN classifiers with regard to overall prediction accuracy, both methods were shown to complement each other, as the sets of true positives, false positives (overprediction), true negatives, and false negatives (underprediction) produced by the two classifiers were not identical. The theory of SVM and ANN training is briefly reviewed.

Algorithms↗
PubMed2006

Patient gender and radiopharmaceutical tracer is of minor importance for the interpretation of myocardial perfusion images using an artificial neural network.

The purpose of this study was to assess the influence of patient gender and choice of perfusion tracer on computer-based interpretation of myocardial perfusion images. For the image interpretation, an automated method was used based on image processing and artificial neural network techniques. A total of 1000 patients were studied, all referred to the Royal Brompton Hospital in London for myocardial perfusion scintigraphy over a period of 1 year. The patients were randomized to receive either thallium or one of the two technetium tracers, methoxyisobutylisonitrile or tetrofosmin. Artificial neural networks were trained with either mixed gender or gender-specific and mixed tracer or tracer-specific training sets of different sizes. The performance of the networks was assessed in separate test sets, with the interpretation of experienced physicians regarding the presence or absence of fixed or reversible defects in the images as the gold standard. The neural networks trained with large mixed gender training sets were as good as the networks trained with gender-specific data sets. In addition, the neural networks trained with large mixed tracer training sets were as good as or better than the networks trained with tracer-specific data sets. Our results indicate that the influence of patient gender and perfusion tracer are of minor importance for the computer-based interpretation of the myocardial perfusion images. The differences that occur can be compensated for by larger training sets.

Adult↗
PubMedDate not supplied

Unstable Morse Code recognition system with back propagation neural network for person with disabilities.

A Morse code auto-recognition system is limited by stable typing speed and stable typing ratio from long to short intervals. For an unstable Morse code typing pattern, the auto-recognition algorithms in the literature are not good enough for applications. This paper adopted a neural network to recognize unstable Morse codes. From an experiment on a teenager with cerebral palsy, the neural network has an average recognition rate up to 93.2%. The recognition rate from an amputee aged 40, who used a prosthesis for typing, it is 97.2% on average. When we compare this to 99.2% for the recognition rate from a skilled expert, the result is quite promising. The neural network has successfully overcome the difficulty of analysing a severely unstable Morse code time series. Since the human typing speed is quite slow in comparison to signal processing by the computer, it also makes it possible to use a neural network for real-time signal recognition.

Adolescent↗
PubMed1999

Artificial neural networks: a potential role in osteoporosis.

Artificial neural networks are computer software systems that recognize patterns in complex data sets. A recent development in neural computing, multiversion systems (MVS), has led to enhanced analytical power, and this was harnessed to demonstrate the value of risk factors in predicting the result of osteoporosis investigations by quantitative ultrasound. 274 women were screened in an open-access osteoporosis service. A conventional risk factor questionnaire was completed for each patient by the osteoporosis specialist nurse. An MVS was trained on 180 randomly selected data sets and tested on the remaining 94. The results were compared with those from logistic regression analysis in predictive power, both from the selected 20-item questionnaire and for a limited 5-item questionnaire comprising age, height, height loss, weight and years since the menopause. The MVS approach predicted the T-score categorization of the patients from the 20-item questionnaire with 83.0% accuracy, whereas logistic regression yielded an accuracy of only 72.8% (P = 0.04). From the 5-item database the MVS yielded a best prediction accuracy of 73.1%, whereas the logistic regression prediction accuracy was 60% (P = 0.04). These results suggest that 20 risk factors can be used by an MVS to predict the outcome of osteoporosis investigations with a power that outperforms conventional statistical methods. Use of this system may improve the selection of patients for osteoporosis investigations, since even with only 5 risk factors the system performs nearly as well as that based on the full 20 factors.

Aged↗
PubMed2000

Feedback error learning neural network for trans-femoral prosthesis.

Feedback-error learning (FEL) neural network was developed for control of a powered trans-femoral prosthesis. Nonlinearities and time-variations of the dynamics of the plant, in addition to redundancy and dynamic uncertainty during the double support phase of walking, makes conventional control methods very difficult to use. Rule-based control, which uses a knowledge base determined by machine learning and finite automata method is limited since it does not respond well to perturbations and environmental changes. FEL can be regarded as a hybrid control, because it combines nonparametric identification with parametric modeling and control. This paper presents simulation of a powered trans-femoral prosthesis controlled by a FEL neural network. Results suggest that FEL can be used to identify inverse dynamics of an arbitrary trans-femoral prosthesis during simple single joint movements (e.g., sinusoidal oscillations). The identified inverse dynamics then allows the tracking of an arbitrary trajectory such as a desired walking pattern within a multijoint structure. Simulation shows that the identified controller responds correctly when the leg motion is exposed to a perturbation such as a frequent change of the ground reaction force or the hip joint torque generated by the user. FEL eliminates the need for precise, tedious, and complex identification of model parameters.

Activities of Daily Living↗
PubMed1997

Stability analysis of a class of noise perturbed neural networks.

We establish robustness stability results for a specific type of artificial neural networks for associative memories under parameter perturbations and determine conditions that ensure the existence of asymptotically stable equilibria of the perturbed neural system that are the asymptotically stable equilibria of the original unperturbed neural network. The proposed stability analysis tool is the sliding mode control and it facilitates the analysis by considering only a reduced-order system instead of the original one and time-dependent external stimuli.

Association Learning↗
PubMed2001

Using a neural network to screen a population for asthma.

PURPOSE: To use a neural network to rank a population according to individual likelihood of asthma based on their responses to a respiratory questionnaire. METHODS: A final diagnosis of asthma can be made only after full clinical assessment but limited resources make it impossible to offer this to complete populations as part of a screening programme. Prioritisation is required so that review can be offered most promptly to those most in need. A stratified random sample of 180 from 6825 respondents to a community survey underwent clinical review. They were categorised according to likelihood of asthma by three independent experts whose opinions were combined into a single probability label for each patient. A neural network was trained to relate questionnaire responses to probability labels. The trained network was applied to the whole community to produce a ranking order based on likelihood of asthma. A screening threshold could then be set to correspond to available resources, and patients above this level with no recorded evidence of asthma diagnosis could be assessed clinically. Using the known probability labels from the training set, it was possible to derive the expected proportion of true asthmatics in any set of patients. RESULTS: If the screening threshold had been set to capture the top 10% of the ranked population (n = 683), then 239 patients above this threshold had no evidence of diagnosis and would need assessment. Of these, it would be expected that 74% would have the diagnosis confirmed. CONCLUSIONS: This approach allows prioritisation of a population where resources for diagnostic examination are limited.

Adolescent↗
PubMed1990

The application of neural networks to myoelectric signal analysis: a preliminary study.

Two neural network implementations are applied to myoelectric signal (MES) analysis tasks. The motivation behind this research is to explore more reliable methods of deriving control for multidegree of freedom arm prostheses. A discrete Hopfield network is used to calculate the time series parameters for a moving average MES model. It is demonstrated that the Hopfield network is capable of generating the same time series parameters as those produced by the conventional sequential least squares (SLS) algorithm. Furthermore, it can be extended to applications utilizing larger amounts of data, and possibly to higher order time series models, without significant degradation in computational efficiency. The second neural network implementation involves using a two-layer perceptron for classifying a single site MES based on two features, specifically the first time series parameter, and the signal power. Using these features, the perceptron is trained to distinguish between four separate arm functions. The two-dimensional decision boundaries used by the perceptron classifier are delineated. It is also demonstrated that the perceptron is able to rapidly compensate for variations when new data are incorporated into the training set. This adaptive quality suggests that perceptrons may provide a useful tool for future MES analysis.

Algorithms↗
PubMed2004

Can we improve the prediction of stone-free status after extracorporeal shock wave lithotripsy for ureteral stones? A neural network or a statistical model?

PURPOSE: We evaluated whether an artificial neural network (ANN) can improve the prediction of stone-free status after extracorporeal shock wave lithotripsy (ESWL) (Dornier Medical Systems, Inc., Marietta, Georgia) for ureteral stones compared to a logistic regression (LR) model. MATERIALS AND METHODS: Between February 1989 and December 1998, 984 patients with ureteral stones, including 780 males and 204 females with a mean age +/- SD of 40.85 +/- 10.33 years, were treated with ESWL. Stone-free status at 3 months was determined by urinary tract plain x-ray and excretory urography. Of all patients 919 (93.3%) were free of stones. The impact of 10 factors on stone-free status was studied using an LR model and ANN. These factors were patient age and sex, renal anatomy, stone location, side, number, length and width, whether stones were de novo or recurrent, and stent use. An LR model was constructed and ANN was trained on 688 randomly selected patients (70%) to predict stone-free status at 3 months. The 10 factors were used as covariates in the LR model and as input parameters to ANN. Performance of the trained net and developed logistic model was evaluated in the remaining 296 patients (30%), who served as the test set. The sensitivity (percent of correctly predicted stone-free cases), specificity (percent of correctly predicted nonstonefree cases), positive predictive value, overall accuracy and average classification rate of the 2 techniques were compared. Relevant variables influencing the construction of the 2 models were compared. RESULTS: Evaluating the performance of the LR and ANN models on the test set revealed a sensitivity of 100% and 77.9%, a specificity of 0.0% and 75%, a positive predictive value of 93.2% and 97.2%, an overall accuracy of 93.2% and 77.7%, and an average classification rate of 50% and 76.5%, respectively. LR failed to predict any nonstone free cases. LR and ANN identified stone location and stent use as important factors in determining the outcome, while ANN also identified stone length and width as influential factors. CONCLUSIONS: ANN and LR could predict adequately those who would be stone-free after ESWL for ureteral stones. The neural network has a higher ability to predict those who fail to respond to ESWL.

Female↗