Antonio J. Serrano

BELM: Bayesian Extreme Learning Machine

The theory of extreme learning machine (ELM) has become very popular on the last few years. ELM is a new approach for learning the parameters of the hidden layers of a multilayer neural network (as the multilayer perceptron or the radial basis function neural network). Its main advantage is the lower computational cost, which is especially relevant when dealing with many patterns defined in a high-dimensional space. This brief proposes a Bayesian approach to ELM, which presents some advantages over other approaches: it allows the introduction of a priori knowledge; obtains the confidence intervals (CIs) without the need of applying methods that are computationally intensive, e.g., bootstrap; and presents high generalization capabilities. Bayesian ELM is benchmarked against classical ELM in several artificial and real datasets that are widely used for the evaluation of machine learning algorithms. Achieved results show that the proposed approach produces a competitive accuracy with some additional advantages, namely, automatic production of CIs, reduction of probability of model overfitting, and use of a priori knowledge.

Neural networks for animal science applications: Two case studies

Abstract Artificial neural networks have shown to be a powerful tool for system modelling in a wide range of applications. In this paper, we focus on neural network applications to intelligent data analysis in the field of animal science. Two classical applications of neural networks are proposed: time series prediction and clustering. The first task is related to the prediction of weekly milk production in goat flocks, which includes a knowledge discovery stage in order to analyse the relative relevance of the different variables. The second task is the clustering of goat flocks; it is used to analyse different livestock surveys by using self-organizing maps and the adaptive resonance theory, thus obtaining a qualitative knowledge from these surveys. Achieved results show the usefulness of neural networks in two animal science applications.

Expert system for predicting unstable angina based on Bayesian networks

The use of computer-based clinical decision support (CDS) tools is growing significantly in recent years. These tools help reduce waiting lists, minimise patient risks and, at the same time, optimise the cost health resources. In this paper, we present a CDS application that predicts the probability of having unstable angina based on clinical data. Due to the characteristics of the variables (mostly binary) a Bayesian network model was chosen to support the system. Bayesian-network model was constructed using a population of 1164 patients, and subsequently was validated with a population of 103 patients. The validation results, with a negative predictive value (NPV) of 91%, demonstrate its applicability to help clinicians. The final model was implemented as a web application that is currently been validated by clinician specialists.

Feature selection using ROC curves on classification problems

Feature Selection (FS) is one of the key stages in classification problems. This paper proposes the use of the area under Receiver Operator Characteristic curves to measure the individual importance of every input as well as a method to discover the variables that yield a statistically significant improvement in the discrimination power of the classification model.

Application of ARMA modeling to the improvement of weight estimations in fruit sorting and grading machinery

Accurate weighting of pieces in different sorts of conveyor belts or articulated chains at fast speed is a key feature in many industrial processes. This paper presents a procedure to improve the performance, whether increasing speed or accuracy, of the load-cell-based weighting subsystem in a fruit sorting and grading machine to achieve an accuracy of /spl plusmn/1 gram at a speed of 20 fruits per seconds. The proposed solution includes a signal preprocessing based on a previous ARMA modeling of the weighting subsystem response plus a power line-noise removal and a simple sample averaging in the plateau. The procedure has been tested off-line using real signals acquired from a prototype machine.

The use of neural networks for predicting the result of endoscopic treatment for vesico‐ureteric reflux

To create an artificial neural network (ANN) to aid in predicting the results of endoscopic treatment for vesico‐ureteric reflux (VUR).

Neural networks as effective techniques in clinical management of patients: some case studies

In this paper, we present four examples of effective implementation of neural systems in the daily clinical practice. There are two main goals in this work; the first one is to show that neural networks are especially well-suited tools for solving different kind of medical/pharmaceutical problems, given the complex input output relationships and the few a priori knowledge about data distribution and variable relations. The second goal is to develop specific software applications, which enclose complex mathematical models, to clinicians; thus, the use of such models as decision support systems is facilitated. Four important pharmaceutical problems are considered in this study: identification of patients with potential risk of postchemotherapy emesis, classification of patients depending on their risk of digoxin intoxication, prediction of cyclosporine A through concentration and prediction of erythropoietin blood concentrations. The Multilayer Perceptron in classification problems and dynamic neural networks, such as the Elman recurrent neural network and the Finite Impulse Response neural network in prediction problems, have been used. Moreover, network ensembles of different kind of networks have been taken into account. Results show that neural networks are suitable tools for medical classification and prediction tasks, outperforming the mostly used methods in these problems (logistic regression and multivariate analysis).

Predicting Early Transplant Failure: Neural Network Versus Logistic Regression Models

Coxs proportional hazard model or logistic regression model has been the classical mathematical approach to predict transplant results, but artificial neural networks may offer better results. In order to compare both methods, a logis- tic regression and a neural network model were generated to predict early transplant failure assessed at 90 days. Methods: Medical charts from 701 liver transplant patients were used as generation cohort, collecting variables from do- nor, recipient and operative data. The discrimination capacity of the models was measured through the area under their ROC curves. Models were validated by applying them to a second cohort of 170 patients (validation cohort), although af- terwards it was enlarged to 246 patients in order to increase statistical power. Results: For the generation sample, ROC curves were 75% for logistic regression and 96% for neural network (� 2 = 44,60. p<0,00001). Applied to the whole validation sample these values dropped to 68.7 % for logistic regression and 69.9 % for neural network (� 2 = 0.026. p: 0,87). However, when models where applied to the validation cohort in cumulative groups of 50 patients two aspects became evident: 1) predictions worsened for patients who were more distant in time from the generation cohort; 2) for the first hundred patients in validation cohort, neural network was clearly superior to logistic re- gression model (93 % vs 76 %; � 2 = 10.52. p:0,001). Conclusions: Our results suggest that, provided with the same information and for a limited period of time, neural net- works may offer better diagnostic performances than with logistic regression models.

Neural Networks Ensemble for Cyclosporine Concentration Monitoring

This paper proposes the use of neural networks ensemble for predicting the cyclosporine A (CyA)concen tration in kidney transplant patients. In order to optimize clinical outcomes and to reduce the cost associated with patient care, accurate prediction of CyA concentrations is the main objective of therapeutic drug monitoring. Thirty-two renal allograft patients and different factors (age, weight, gender, creatinine and post-transplantation days, together with past dosages and concentrations)w ere studied to obtain the best models. Three kinds of networks (multilayer perceptron, FIR network, Elman recurrent network) and the formation of neural-network ensembles were used. The FIR network, yielding root-mean-squared errors (RMSE)of 41.61 ng/mL in training (22 patients)and 52.34 ng/mL in validation (10 patients)sho wed the best results. A committee of trained networks improved accuracy (RMSE = 44.77 ng/mL in validation).

NEMESIS: a new telemedicine approach to cardiologic software

The present communication describes a telemedicine approach to a computer-aided healthcare system. The application is a comprehensive tool set, structured around a main navigation bar, which leads to every sub-application. Those sub-applications are window based user-friendly tools, comprising specific health tools for physicians, telemedicine/telematics tools and general-purpose tools. The whole application is built in a highly structured and modular way, which allows adding or excluding modules quickly. Portability is guaranteed and achieved by Java programming language, bringing independent platform software. All modules comply with international standards; namely, medical modules comply with European CEN TC 251 approved standards, while telematic modules observe the ITU standards. The present application release offers one specific medical module, videoconference, electronic mail, remote control, electronic healthcare patient record retrieval and knowledge database access.