Training a learning vector quantization network using the pattern electroretinography signals

In this study, the pattern electroretinography (PERG) signals derived from evoked potential across retinal cells of subjects after visual stimulation were analyzed using artificial neural network (ANN) with 172 healthy and 148 diseased subjects. ANN was employed to PERG signals to distinguish between healthy eye and diseased eye. Supervised network examined was a competitive learning vector quantization network. The designed classification structure has about 94% sensitivity, 90.32% specifity, 5.94% false negative, 9.67% false positive and correct classification is calculated to be 92%. Testing results were found to be compliant with the expected results that are derived from the physician's direct diagnosis. The end benefit would be to assist the physician to make the final decision without hesitation.     

A fuzzy clustering neural network architecture for classification of ECG arrhythmias

Accurate and computationally efficient means of classifying electrocardiography (ECG) arrhythmias has been the subject of considerable research effort in recent years. This study presents a comparative study of the classification accuracy of ECG signals using a well-known neural network architecture named multi-layered perceptron (MLP) with backpropagation training algorithm, and a new fuzzy clustering NN architecture (FCNN) for early diagnosis. The ECG signals are taken from MIT-BIH ECG database, which are used to classify 10 different arrhythmias for training. These are normal sinus rhythm, sinus bradycardia, ventricular tachycardia, sinus arrhythmia, atrial premature contraction, paced beat, right bundle branch block, left bundle branch block, atrial fibrillation and atrial flutter. For testing, the proposed structures were trained by backpropagation algorithm. Both of them tested using experimental ECG records of 92 patients (40 male and 52 female, average age is 39.75 +/- 19.06). The test results suggest that a new proposed FCNN architecture can generalize better than ordinary MLP architecture and also learn better and faster. The advantage of proposed structure is a result of decreasing the number of segments by grouping similar segments in training data with fuzzy c-means clustering.     

Potential of the Genetic Algorithm Neural Network in the Assessment of Gait Patterns in Ankle Arthrodesis

The aim of this study was to develop an empirical model of parameter-based gait data, based on an artificial neural network and a genetic algorithm, for the assessment of patients after ankle arthrodesis. Ground reaction force vectors were measured by force platforms during level walking. Nine force parameters expressed in percentage of body weight and their chronologic incidence of occurrence expressed in percentage of stance phase period were used in modeling. Ten healthy persons and ten patients who had solid arthrodesis of the ankle were recruited in this study for developing the model. By applying the genetic algorithm neural network, the percentage of correct classification was 98.8% and the subset of discriminant parameters was be reduced to 9 out of 18. These key parameters were mainly related to the loading response and propulsive phase. This indicates that there was a reduction in the abilities in cushion impact and push off in the patients after ankle arthrodesis. Finally, the relative distance (D _r ) was defined in this study and used in two new patients' examinations to demonstrate its clinical utility. © 2001 Biomedical Engineering Society. PAC01: 8719St, 0705Mh, 8780-y, 8710+e     

Artificial neural network analysis of gustatory responses in the thalamic taste relay of the rat

We used an artificial neural network (ANN) as a model for analyzing single-neuron responses from the thalamic taste relay of rats. The network consisted of: (1) a layer of 44 input units, representing the responses of the 44 thalamic taste cells; (2) a layer of hidden units of varying numbers; and (3) a layer of four output units. We used the back-propagation algorithm to train the output units to discriminate among tastants based on inputs from the thalamic neurons. As the network became fully trained, we found that: (1) only two hidden units were necessary to provide nearly the full discriminative capacity of the network; (2) the loss of even a few of the input units that had the highest impact on hidden units caused a drastic reduction of discriminative power, implying that not all neurons contribute equally to the neural code; and (3) adding a temporal component to the input, by representing each 100-ms time bin as a separate input unit, increased the accuracy with which output units were able to identify tastants. We used data from behavioral discrimination tasks as a measure of the capacity of the network to identify stimuli correctly. A network with two hidden units was about as effective as an across-pattern analysis in accounting for the rat's discriminative ability.     

Hierarchical state space partitioning with a network self-organising map for the recognition of ST-T segment changes

The problem of maximising the performance of ST-T segment automatic recognition for ischaemia detection is a difficult pattern classification problem. The paper proposes the network self-organising map (NetSOM) model as an enhancement to the Kohonen self-organised map (SOM) model. This model is capable of effectively decomposing complex large-scale pattern classification problems into a number of partitions, each of which is more manageable with a local classification device. The NetSOM attempts to generalise the regularisation and ordering potential of the basic SOM from the space of vectors to the space of approximating functions. It becomes a device for the ordering of local experts (i.e. independent neural networks) over its lattice of neurons and for their selection and co-ordination. Each local expert is an independent neural network that is trained and activated under the control of the NetSOM. This method is evaluated with examples from the European ST-T database. The first results obtained after the application of NetSOM to ST-T segment change recognition show a significant improvement in the performance compared with that obtained with monolithic approaches, i.e. with single network types. The basic SOM model has attained an average ischaemic beat sensitivity of 73.6% and an average ischaemic beat predictivity of 68.3%. The work reports and discusses the improvements that have been obtained from the implementation of a NetSOM classification system with both multilayer perceptrons and radial basis function (RBF) networks as local experts for the ST-T segment change problem. Specifically, the NetSOM with multilayer perceptrons (radial basis functions) as local experts has improved the results over the basic SOM to an average ischaemic beat sensitivity of 75.9% (77.7%) and an average ischaemic beat predictivity of 72.5% (74.1%).     

Recognition of the physiological actions of the triphasic EMG pattern by a dynamic recurrent neural network

Triphasic electromyographic (EMG) patterns with a sequence of activity in agonist (AG1), antagonist (ANT) and again in agonist (AG2) muscles are characteristic of ballistic movements. They have been studied in terms of rectangular pulse-width or pulse-height modulation. In order to take into account the complexity of the EMG signal within the bursts, we used a dynamic recurrent neural network (DRNN) for the identification of this pattern in subjects performing fast elbow flexion movements. Biceps and triceps EMGs were fed to all 35 fully-connected hidden units of the DRNN for mapping onto elbow angular acceleration signals. DRNN training was supervised, involving learning rule adaptations of synaptic weights and time constants of each unit. We demonstrated that the DRNN is able to perfectly reproduce the acceleration profile of the ballistic movements. Then we tested the physiological plausibility of all the networks that reached an error level below 0.001 by selectively increasing the amplitude of each burst of the triphasic pattern and evaluating the effects on the simulated accelerating profile. Nineteen percent of these simulations reproduced the physiological action classically attributed to the 3 EMG bursts: AG1 increase showed an increase of the first accelerating pulse, ANT an increase of the braking pulse and AG2 an increase of the clamping pulse. These networks also recognized the physiological function of the time interval between AG1 and ANT, reproducing the linear relationship between time interval and movement amplitude. This task-dynamics recognition has implications for the development of DRNN as diagnostic tools and prosthetic controllers.     

Assessing the user experience of older adults using a neural network trained to recognize emotions from brain signals

The use of Ambient Assisted Living (AAL) technologies as a means to cope with problems that arise due to an increasing and aging population is becoming usual. AAL technologies are used to prevent, cure and improve the wellness and health conditions of the elderly. However, their adoption and use by older adults is still a major challenge. User Experience (UX) evaluations aim at aiding on this task, by identifying the experience that a user has while interacting with an AAL technology under particular conditions. This may help designing better products and improve user engagement and adoption of AAL solutions. However, evaluating the UX of AAL technologies is a difficult task, due to the inherent limitations of their subjects and of the evaluation methods. In this study, we validated the feasibility of assessing the UX of older adults while they use a cognitive stimulation application using a neural network trained to recognize pleasant and unpleasant emotions from electroencephalography (EEG) signals by contrasting our results with those of additional self-report and qualitative analysis UX evaluations. Our study results provide evidence about the feasibility of assessing the UX of older adults using a neural network that take as input the EEG signals; the classification accuracy of our neural network ranges from 60.87% to 82.61%. As future work we will conduct additional UX evaluation studies using the three different methods, in order to appropriately validate these results.     

A novel application of pattern recognition for accurate SNP and indel discovery from high-throughput data: targeted resequencing of the glucocorticoid receptor co-chaperone FKBP5 in a Caucasian population

The detection of single nucleotide polymorphisms (SNPs) and insertion/deletions (indels) with precision from high-throughput data remains a significant bioinformatics challenge. Accurate detection is necessary before next-generation sequencing can routinely be used in the clinic. In research, scientific advances are inhibited by gaps in data, exemplified by the underrepresented discovery of rare variants, variants in non-coding regions and indels. The continued presence of false positives and false negatives prevents full automation and requires additional manual verification steps. Our methodology presents applications of both pattern recognition and sensitivity analysis to eliminate false positives and aid in the detection of SNP/indel loci and genotypes from high-throughput data. We chose FK506-binding protein 51(FKBP5) (6p21.31) for our clinical target because of its role in modulating pharmacological responses to physiological and synthetic glucocorticoids and because of the complexity of the genomic region. We detected genetic variation across a 160 kb region encompassing FKBP5. 613 SNPs and 57 indels, including a 3.3 kb deletion were discovered. We validated our method using three independent data sets and, with Sanger sequencing and Affymetrix and Illumina microarrays, achieved 99% concordance. Furthermore we were able to detect 267 novel rare variants and assess linkage disequilibrium. Our results showed both a sensitivity and specificity of 98%, indicating near perfect classification between true and false variants. The process is scalable and amenable to automation, with the downstream filters taking only 1.5 h to analyze 96 individuals simultaneously. We provide examples of how our level of precision uncovered the interactions of multiple loci, their predicted influences on mRNA stability, perturbations of the hsp90 binding site, and individual variation in FKBP5 expression. Finally we show how our discovery of rare variants may change current conceptions of evolution at this locus.     

Tracking the states of a nonlinear and nonstationary system in the weight-space of artificial neural networks

We propose a novel interpretation and usage of Neural Network (NN) in modeling physiological signals, which are allowed to be nonlinear and/or nonstationary. The method consists of training a NN for the k-step prediction of a physiological signal, and then examining the connection-weight-space (CWS) of the NN to extract information about the signal generator mechanism. We define a novel feature, Normalized Vector Separation (γ _ij ), to measure the separation of two arbitrary states “i” and “j” in the CWS and use it to track the state changes of the generating system. The performance of the method is examined via synthetic signals and clinical EEG. Synthetic data indicates that γ _ij can track the system down to a SNR of 3.5 dB. Clinical data obtained from three patients undergoing carotid endarterectomy of the brain showed that EEG could be modeled (within a root-means-squared-error of 0.01) by the proposed method, and the blood perfusion state of the brain could be monitored via γ _ij , with small NNs having no more than 21 connection weight altogether.     

Estimation of pulmonary arterial pressure by a neural network analysis using features based on time-frequency representations of the second heart sound

The objective of the study was to develop a non-invasive method for the estimation of pulmonary arterial pressure (PAP) using a neural network (NN) and features extracted from the second heart sound (S2). To obtain the information required to train and test the NN, an animal model of pulmonary hypertension (PHT) was developed, and nine pigs were investigated. During the experiments, the electrocardiogram, phonocardiogram and PAP were recorded. Subsequently, between 15 and 50 S2 heart sounds were isolated for each PAP stage and for each animal studied. A Coiflet wavelet decomposition and a pseudo smoothed Wigner-Ville distribution were used to extract features from the S2 sounds and train a one-hidden-layer NN using two-thirds of the data. The NN performance was tested on the remaining one-third of the data. NN estimates of the systolic and mean PAPs were obtained for each S2 and then ensemble averaged over the 15–50 S2 sounds selected for each PAP stage. The standard errors between the mean and systolic PAPs estimated by the NN and those measured with a catheter were 6.0 mmHg and 8.4 mmHg, respectively, and the correlation coefficients were 0.89 and 0.86, respectively. The classification accuracy, using 23 mmHg mean PAP and 30 mmHg systolic PAP thresholds between normal PAP and PHT, was 97% and 91% respectively.     

Lesion of areas 17/18/19: effects on the cat's performance in a binary detection task

Summary The ability of two cats to discriminate between two geometrical outline patterns in the presence of superimposed Gaussian visual noise — i.e. in a binary detection task — was tested before and after bilateral removal of cortical areas 17, 18 and 19. The detection probability P_D was measured as a function of the signal-to-noise ratio. After a lesion of areas 17, 18 and 19 both cats were unable to carry out the discrimination tasks. Their detection performance dropped to chance level, but after an extensive phase of retraining (3 months) they regained the ability to discriminate visual patterns. It was thus possible to obtain detection curves and to determine a measure of a performance which is predominantly bound to be mediated by extra-geniculo-cortical systems. The detection capacity was abnormally low with both large and small patterns. However, the detection of stationary small patterns was similar to the performance of cats with 17/18 lesions; the detection of stationary large patterns was only slightly better than the detection of small patterns and much worse than the comparable performance of cats with 17/18 lesions. Furthermore the cats with lesions of areas 17/18/19 were unable to discriminate moving patterns, their performances being at chance level, whereas for the cats with 17/18 lesions the detection of moving and stationary patterns was equal.Supported by the Deutsche Forschungsgesellschaft     

Automatic detection method of P300 waveform in the single sweep records by using a neural network

Event-related potentials (ERPs) are responses related to the recognition of certain stimuli. P300 is the most important positive component in the ERP and appears around 300 ms after the target stimulus in the oddball paradigm. In our previous work, we proposed a method for the automatic detection of the P300 waveform in single-sweep records by using a correlation technique. However, determination of the threshold values of the P300 waveform for the correlation study was not an easy task. In skirting this problem, we developed an automatic method of detecting a single-sweep P300 waveform by using an artificial neural network. We selected appropriate characteristic parameters of positive peaks as input signals for the input layer units, and the weights between the layers were determined by using the back-propagation algorithm. The neural network for P300 detection was obtained automatically, based on the data of ERPs obtained from 11 healthy males, and gave substantial accuracy for P300 detection. Furthermore, by using this neural network we clarified the way in which the P300 waveform is judged visually by each inspector.     

The use of a neural network to forecast daily grass pollen concentration in a Mediterranean region: the southern part of the Iberian Peninsula

BACKGROUND: Pollen allergy is a common disease causing hayfever in 15% of the population in Europe. Medical studies report that a prior knowledge of pollen content in the air can be useful in the management of pollen-related diseases. OBJECTIVES: The aim of this work was to forecast daily Poaceae pollen concentrations in the air by using meteorological data and pollen counts from previous days as independent variables. METHODS: Linear regression models and co-evolutive neural network models were used for this study. Pollen was monitored by a Hirst-type spore trap using standard techniques. The data were obtained from the Spanish Aerobiology Network database, University of Cordoba Monitoring Unit. The set of data includes a series of 20 years, from 1982 to 2001. A classification of the years according to their allergenic potential was made using a K-mean cluster analysis with pollen and meteorological parameters. Statistical analysis was applied to all the years of each class with the exception of the most recent year, which was used for model validation. RESULTS: It was observed that cumulative variables and pollen values from previous days are the most important factors in the models. In general, neural network equations produce better results than linear regression equations. CONCLUSION: Co-evolutive neural network models, which obtain the best forecasts (an almost 90% "good" classification), make it possible to predict daily airborne Poaceae pollen concentrations. This new system based on neural network models is a step toward the automation of the pollen forecast process.     

Evaluation of an artificial neural network to ascertain why there is a high incidence of hepatitis B in the Chinese population after vaccination

Objective

The purpose of the study was to understand why there is a high incidence of hepatitis B, including acute and chronic infections, after vaccination.

Methods

All data were obtained from the published information. The incidence of hepatitis B, life expectancy, and positive rates of hepatitis B surface antigen (HBsAg) and the antibody to HBsAg (anti-HBs) in a specific observation year was depicted. The relationships between the HBsAg-positive rate, anti-HBs-positive rate, life expectancy and incidence of hepatitis B were assessed by Pearson’s correlation. The quantitative effects of HBsAg-positive rate, anti-HBs-positive rate and life expectancy on hepatitis B incidence were assessed by SPSS 18.0 software with a neural network.

Results

There was no correlation between hepatitis B incidence and the rates of HBsAg or anti-HBs (r=−0.334 for HBsAg, r=0.247 for anti-HBs, p>0.05 for both indicators). A clear correlation was observed between incidence and life expectancy (r=0.841, p<0.05). The standardized percentage of affect of life expectancy on hepatitis B incidence (100%) was much higher than that of either the rates of HBsAg (59.0%) or anti-HBs (45.7%).

Conclusion

An increase in life expectancy and an improved quality of life may be associated with a higher incidence of hepatitis B. Because hepatitis B is caused by the host immune response, a stronger immune response in individuals with an improved quality of life could contribute to that higher incidence.