Data classification with multilayer perceptrons using a generalized error function.

The learning process of a multilayer perceptron requires the optimization of an error function E(y,t) comparing the predicted output, y, and the observed target, t. We review some usual error functions, analyze their mathematical properties for data classification purposes, and introduce a new one, E(Exp), inspired by the Z-EDM algorithm that we have recently proposed. An important property of E(Exp) is its ability to emulate the behavior of other error functions by the sole adjustment of a real-valued parameter. In other words, E(Exp) is a sort of generalized error function embodying complementary features of other functions. The experimental results show that the flexibility of the new, generalized, error function allows one to obtain the best results achievable with the other functions with a performance improvement in some cases.

A novel dry active electrode for EEG recording.

The design and testing of a "dry" active electrode for electroencephalographic recording is described. A comparative study between the EEG signals recorded in human volunteers simultaneously with the classical Ag-AgCl and "dry" active electrodes was carried out and the reported preliminary results are consistent with a better performance of these devices over the conventional Ag-AgCl electrodes.

Error entropy in classification problems: a univariate data analysis.

Entropy-based cost functions are enjoying a growing attractiveness in unsupervised and supervised classification tasks. Better performances in terms both of error rate and speed of convergence have been reported. In this letter, we study the principle of error entropy minimization (EEM) from a theoretical point of view. We use Shannon's entropy and study univariate data splitting in two-class problems. In this setting, the error variable is a discrete random variable, leading to a not too complicated mathematical analysis of the error entropy. We start by showing that for uniformly distributed data, there is equivalence between the EEM split and the optimal classifier. In a more general setting, we prove the necessary conditions for this equivalence and show the existence of class configurations where the optimal classifier corresponds to maximum error entropy. The presented theoretical results provide practical guidelines that are illustrated with a set of experiments with both real and simulated data sets, where the effectiveness of EEM is compared with the usual mean square error minimization.

SisPorto 2.0: a program for automated analysis of cardiotocograms.

OBJECTIVE: To describe the latest version of SisPorto, a program for automated analysis of cardiotocograms that closely follows the FIGO guidelines, analyses ante- and intrapartum tracings, performs no signal reduction, and has the possibility of simultaneously recording twins. METHODS: A detailed description of the program's processing algorithms and operation is provided, as well as the main results of the studies performed to-date with this system. RESULTS: Considering both current and previous versions of the program, SisPorto has been tested in over 6000 pregnancies. The system's FHR baseline was compared with an average of three experts' estimates, and the difference was under 8 bpm in all cases. A fair to good agreement was found with experts' identification of accelerations, decelerations, contractions, and normal/reduced variability (proportions of agreement 0.64-0.89). In a preliminary validity study (n = 85), a sensitivity of 100% and a specificity of 99% were obtained in prediction of poor neonatal outcome. The system is currently undergoing an international multicentre validation study. CONCLUSIONS: Although still at the research level, a considerable experience has now been gathered with this system. Promising results have been achieved in studies comparing SisPorto with experts' analysis and in those evaluating the validity of the system.

Frontal-plane QRS axis revisited: accuracy of current approximations and reappraisal of their merit in the diagnosis of right ventricular hypertrophy.

The frontal-plane mean QRS vector orientation (AQRSxy)--the so-called electrical axis--is an ECG feature commonly used for the diagnosis of right ventricular hypertrophy and is correctly measured by calculating the areas subtended by QRS deflections in two different leads. To overcome the drawbacks of doing this by hand, two alleged approximations of AQRSxy have become popular and are in current use: one is based on the measurement of QRS component wave peak amplitudes and the other on the estimation of the half-area vector of the frontal plane loop. The values obtained with the correct and the two more practical methods are compared and their diagnostic efficiency is assessed by means of a procedure for ECG criteria optimization based on the receiver operating characteristics (ROC) curve analyzed in terms of information theory. The authors conclude that the two more popular methods for AQRSxy determination provide similar values that, although correlated with the true measure of the parameter are statistically different from it. On the other hand, the diagnostic efficiency of AQRSxy alone, regardless of the method by which it is computed, is only as good as, if not bettered by, other much more easily measurable frontal-plane parameters (ie, left to rightward forces amplitude ratio in adults and rightward forces amplitude in pediatric patients).

A new ECG classifier based on linear prediction techniques.

A longstanding tradition in automatic ECG classification has been the use of conventional features (amplitudes, duration, etc.) as waveform descriptors for pattern discrimination purposes. This paper presents an alternative approach in statistical ECG classification. It is based on the use of linear prediction coefficients, a sort of "abstract" features which, as waveform descriptors, enjoy the desirable property of whole-signal dependency, being rather insensitive to high-frequency noise. Experimental results obtained on 400 ECGs distributed by four clinical groups according to clinicopathological data (normal, myocardial infarction, right and left hypertrophies) show interesting potentialities of this new method, namely a classification error for equal class prevalences (30%) significantly lower than by using conventional features. Classification and cluster separability results are presented and discussed as well as the viability of the new method in a clinical environment.