Search

e-offprint for personal use only and shall not be selfarchived in electronic repositories. Posture transitions are one of the most mechanically demanding tasks and are useful to evaluate the motor status of patients with motor impairments, frail individuals or the elderly, among others. So far, wearable inertial systems have been one of the most employed tools in the study of these movements due to their suitable size and weight, being non-invasive systems. These devices are mainly composed of accelerometers and, to a lesser extent, gyroscopes, magnetometers or barometers. Although accelerometers provide the most reliable measurement, detecting activities where a change of altitude is observed, such as some posture transitions, may require additional sensors to reliably detect these activities. In this work, we present an algorithm that combines the information of a barometer and an accelerometer to detect posture transitions and falls. In contrast to other works, we test different activities (where altitude is involved) in order to achieve a reliable classifier against false positives. Furthermore, by means of feature selection methods, we obtain optimal subsets of features for the accelerometer and barometer sensors to contextualise these activities. The selected features are tested through several machine learning classifiers, which are assessed with an evaluation data set. Results show that the inclusion of barometer features in addition to those obtained for an accelerometer clearly enhances the detection accuracy up to a 11%, in terms of geometric mean between sensitivity and specificity, compared to algorithms where only the accelerometer is used. Finally, we have also analysed the computer burden; in this sense, the usage of barometers, in addition to increase the accuracy, also reduces the computational resources required to classify a new pattern, as shown by a reduction in the number of support vectors.

Among Parkinson’s disease (PD) motor symptoms, freezing of gait (FOG) may be the most incapacitating. FOG episodes may result in falls and reduce patients’ quality of life. Accurate assessment of FOG would provide objective information to neurologists about the patient’s condition and the symptom’s characteristics, while it could enable non-pharmacologic support based on rhythmic cues. This paper is, to the best of our knowledge, the first study to propose a deep learning method for detecting FOG episodes in PD patients. This model is trained using a novel spectral data representation strategy which considers information from both the previous and current signal windows. Our approach was evaluated using data collected by a waist-placed inertial measurement unit from 21 PD patients who manifested FOG episodes. These data were also employed to reproduce the state-of-the-art methodologies, which served to perform a comparative study to our FOG monitoring system. The results of this study demonstrate that our approach successfully outperforms the state-of-the-art methods for automatic FOG detection. Precisely, the deep learning model achieved 90% for the geometric mean between sensitivity and specificity, whereas the state-of-the-art methods were unable to surpass the 83% for the same metric.

Among Parkinson's disease (PD) symptoms, freezing of gait (FoG) is one of the most debilitating. To assess FoG, current clinical practice mostly employs repeated evaluations over weeks and months based on questionnaires, which may not accurately map the severity of this symptom. The use of a non-invasive system to monitor the activities of daily living (ADL) and the PD symptoms experienced by patients throughout the day could provide a more accurate and objective evaluation of FoG in order to better understand the evolution of the disease and allow for a more informed decision-making process in making adjustments to the patient's treatment plan. This paper presents a new algorithm to detect FoG with a machine learning approach based on Support Vector Machines (SVM) and a single tri-axial accelerometer worn at the waist. The method is evaluated through the acceleration signals in an outpatient setting gathered from 21 PD patients at their home and evaluated under two different conditions: first, a generic model is tested by using a leave-one-out approach and, second, a personalised model that also uses part of the dataset from each patient. Results show a significant improvement in the accuracy of the personalised model compared to the generic model, showing enhancement in the specificity and sensitivity geometric mean (GM) of 7.2%. Furthermore, the SVM approach adopted has been compared to the most comprehensive FoG detection method currently in use (referred to as MBFA in this paper). Results of our novel generic method provide an enhancement of 11.2% in the GM compared to the MBFA generic model and, in the case of the personalised model, a 10% of improvement with respect to the MBFA personalised model. Thus, our results show that a machine learning approach can be used to monitor FoG during the daily life of PD patients and, furthermore, personalised models for FoG detection can be used to improve monitoring accuracy. Part of this project has been performed within the framework of the MASPARK project which is funded by La Fundació La Marató de TV3 20140431. This work also forms part of the framework of the FP7 project REMPARK ICT-287677, which is funded by the European Community. peer-reviewed

IWANN is a biennial conference that seeks to provide a discussion forum for scientists, engineers, educators and students about the latest ideas and realizations in the foundations, theory, models and applications of hybrid systems inspired on nature (neural networks, fuzzy logic and evolutionary systems) as well as in emerging areas related to the above items. As in previous editions of IWANN, it also aims to create a friendly environment that could lead to the establishment of scientific collaborations and exchanges among attendees. Since the first edition in Granada (LNCS 540, 1991), the conference has evolved and matured, and most of the topics involved have achieved a maturity and reinforced consolidation. The twelveth edition of the IWANN conference “International Work-Conference on Artificial Neural Networks” was held in Puerto de la Cruz, Tenerife, (Spain) during June 12–14, 2013. The list of topics in the successive Call for Papers has also evolved, resulting in the following list for the present edition

This paper describes a Fault Tolerance System (FTS) implemented in a new self-adaptive hardware architecture. This architecture is based on an array of cells that implements in a distributed way self-adaptive capabilities. The cell includes a configurable multiprocessor, so it can have between one and four processors working in parallel, with a programmable configuration mode that allows selecting the size of program and data memories. The self-elimination and self-replication capabilities of cell(s) are performed when the FTS detects a failure in any of the processors that include it, so that this cell(s) will be self-discarded for future implementations. Other adaptive capabilities of the system are self-routing, self-placement and runtime selfconfiguration. Additionally, it is described as an example application and a software tool that has been implemented to facilitate the development of applications to test the system., Este artículo describe un sistema de tolerancia a fallos (FTS) implementado en una nueva arquitectura de hardware autoadaptativa. Esta arquitectura se basa en una matriz de células que implementa de forma distribuida capacidades autoadaptativas. La célula incluye un multiprocesador configurable, por lo que puede tener entre uno y cuatro procesadores trabajando en paralelo, con un modo de configuración programable que permite seleccionar el tamaño de las memorias de programa y datos. Las capacidades de autoeliminación y autorreplicación de la(s) célula(s) se llevan a cabo cuando el FTS detecta un fallo en alguno de los procesadores que la(s) incluye, de forma que esta(s) célula(s) se autodescarta(n) para futuras implementaciones. Otras capacidades adaptativas del sistema son el autoenrutamiento, la autocolocación y la autoconfiguración en tiempo de ejecución. Además, se describe una aplicación de ejemplo y una herramienta de software que se ha implementado para facilitar el desarrollo de aplicaciones para probar el sistema.

Analysis of human body movement is an important research area, specially for health applications. In order to assess the quality of life of people with mobility problems like Parkinson’s disease o stroke patients, it is crucial to monitor and assess their daily life activities. The main goal of this work is the characterization of basic activities using a single triaxial accelerometer located at the waist. This paper presents a novel postural detection algorithm based in SVM methods which is able to detect and identify Walking, Stand, Sit, Lying, Sit to Stand, Stand to sit, Bending up/down, Lying from Sit and Sit from Lying transitions with a sensitivity of 97% and specificity of 84% with 2884 postures analyzed from 31 healthy volunteers. Parameters and models found have been tested in another dataset from Parkinson’s disease patients, achieving results of 98% of sensitivity and 78% of specificity in postural transitions. The proposed algorithm has been optimized to be easily implemented in real-time system for on-line monitoring applications.

This paper presents a novel gait recognition method which uses the signals measured by a single inertial sensor located on the waist. This method considers human gait as a dynamical system and employs a few singular values obtained by means of Singular Spectrum Analysis applied to scalar measurements from the inertial sensor. Singular values can be interpreted as the approximate edge length of the bounding box wrapping the attractor in the latent space. Effects of different parameters on the gait recognition performance using patterns from 20 different subjects are analysed.

HighlightsThe necessary algorithms to evaluate the occurrence of dopaminergic-induced dyskinesias in the activities of daily life are developed.Sensor placement at the waist provides a good resolution for almost any choreic dyskinesias and provides a good usability and comfort to the patient.A new frequency based approach is proposed to evaluate the occurrence of dyskinesias.The algorithm presented has been evaluated on a database of signals of 92 PD patients and provides specificities and sensitivities above 90%. BackgroundAfter several years of treatment, patients with Parkinson's disease (PD) tend to have, as a side effect of the medication, dyskinesias. Close monitoring may benefit patients by enabling doctors to tailor a personalised medication regimen. Moreover, dyskinesia monitoring can help neurologists make more informed decisions in patient's care. ObjectiveTo design and validate an algorithm able to be embedded into a system that PD patients could wear during their activities of daily living with the purpose of registering the occurrence of dyskinesia in real conditions. Materials and methodsData from an accelerometer positioned in the waist are collected at the patient's home and are annotated by experienced clinicians. Data collection is divided into two parts: a main database gathered from 92 patients used to partially train and to evaluate the algorithms based on a leave-one-out approach and, on the other hand, a second database from 10 patients which have been used to also train a part of the detection algorithm. ResultsResults show that, depending on the severity and location of dyskinesia, specificities and sensitivities higher than 90% are achieved using a leave-one-out methodology. Although mild dyskinesias presented on the limbs are detected with 95% specificity and 39% sensitivity, the most important types of dyskinesia (any strong dyskinesia and trunk mild dyskinesia) are assessed with 95% specificity and 93% sensitivity. ConclusionThe presented algorithmic method and wearable device have been successfully validated in monitoring the occurrence of strong dyskinesias and mild trunk dyskinesias during activities of daily living.

The electrical interfacing with cultured networks of neurons by means of multielectrode arrays (MEAs) is an area of intensive research given the potential of this technique to aid in the extraction of the algorithms that support neurocomputation in brain tissue. To this end, we have previously described the polymer-on-multielectrode technology (PoM [E. Claverol-Tinture, M. Ghirardi, F. Fiumara, X. Rosell, J. Cabestany. Multielectrode arrays with elastomeric microstructures for neuronal patterning towards interfacing with uni-dimensional neuronal networks. J. Neural Eng. 2(2) (2005) L1-7] and demonstrated single-site recordings from patterned invertebrate cells. The realisation of the full potential of PoM is dependent on its successful application to vertebrate cells and to the recording of activity in ensembles of neurons. Here, we describe progress in these directions, specifically towards the development of vertebrate neuronal cultures devoid of glial layers compatible with PoM devices, the connection of pairs of invertebrate neurons threading microchannels and the recordings of synaptic and spike-like activity.

Otoliths have traditionally been used to estimate fish age. However, many factors influence changes in otolith shape, so manual classification remains a complicated task. Very recently, statistical learning techniques have been proposed for automating such a process. We propose performing automatic fish age classification using otolith images (in cases in which growth rings are not properly displayed or are unavailable), morphological and statistical feature-extraction methods and multi-class support vector machines. The results of our experiments, in which we classified cod ages from otolith images, demonstrate the effectiveness of the approach.

Identification of activities of daily living is essential in order to evaluate the quality of life both in the elderly and patients with mobility problems. Posture transitions (PT) are one of the most mechanically demanding activities in daily life and, thus, they can lead to falls in patients with mobility problems. This paper deals with PT recognition in Parkinson's disease (PD) patients by means of a triaxial accelerometer situated between the anterior and the left lateral part of the waist. Since sensor's orientation is susceptible to change during long monitoring periods, a hierarchical structure of classifiers is proposed in order to identify PT while allowing such orientation changes. Results are presented based on signals obtained from 20 PD patients and 67 healthy people who wore an inertial sensor on different positions among the anterior and the left lateral part of the waist. The algorithm has been compared to a previous approach in which only the anterior-lateral location was analyzed improving the sensitivity while preserving specificity. Moreover, different supervised machine learning techniques have been evaluated in distinguishing PT. Results show that the location of the sensor slightly affects method's performance and, furthermore, PD motor state does not alter its accuracy. Posture transition identification is performed by means of a tri-axial accelerometer located in the waist.A hierarchical structure of classifiers allows to determine the human posture.SVM techniques have been used to set parameters of the algorithm.The algorithm allows different locations along waist's left side.The algorithm is focused on Parkinson's disease patients.

Electrochemical sensors, like ion-selective field transistors (ISFET), are electronic devices that merge solid-state electronic technology with chemical sensors so as to be sensitive to the concentration of a particular ion in a solution. However, as it has been previously reported, their response does not only depend on a single ion but also is affected by several interfering ions found in the solution to be measured. These interfering ions can be considered as noise and consequently, a post-processing stage that increases the SNR is obligatory. Our work shows how ensemble learning methods could be used in an array of chemical sensors in order to deal with this problem. In particular, we introduce a novel neural learning architecture for ISFET arrays, which employ ISFET models as prior knowledge. The proposed ensemble learning systems are RBF-like solutions based on bagging and optimal linear combination. Several experimental results are included, which demonstrate the interest and viability of the proposed solution.

Ensemble learning is a well-established method for improving the generalization performance of learning machines. The idea is to combine a number of learning systems that have been trained in the same task. However, since all the members of the ensemble are operating at the same time, large amounts of memory and long execution times are needed, limiting its practical application. This paper presents a new method (called local averaging) in the context of nearest neighbor (NN) classifiers that generates a classifier from the ensemble with the same complexity as the individual members. Once a collection of prototypes is generated from different learning sessions using a Kohonen's LVQ algorithm, a single set of prototypes is computed by applying a cluster algorithm (such as K-means) to this collection. Local averaging can be viewed either as a technique to reduce the variance of the prototypes or as the result of averaging a series of particular bootstrap replicates. Experimental results using several classification problems confirm the utility of the method and show that local averaging can compute a single classifier that achieves a similar (or even better) accuracy than ensembles generated with voting.

The asymptotic convergence of on-line algorithms when the number of training samples becomes infinite is well understood from a theoretical point of view (Adaptive Algorithms and Stochastic Approximations, Springer, Berlin, 1990, Advances in Neural Processing Systems 7, MIT Press, Boston, 1995, Theory and Practice of Recursive Identification, MIT Press, Boston, 1983). However, much less is known about the real convergence of these algorithms when the data sample size is finite. In this paper, we address the study of the real convergence of the popular K-means algorithm (Proceedings of the Fifth Berkeley Symposium on Mathematics, Statistics and Probablity, Vol. 1, 1967, 281) when it deals with finite data resources.

In the present issue of Neurocomputing, it is apleasure to present you a collection of 12 extended versions of selected papers from the 10 the dition of the International Work Conference on Artificial Neural Networks (IWANN2009). This is a conference held every two year in Spain, and focus in gon the foundations, theory, models and applications of systems, which are inspired by nature (e.g.neural networks, fuzzy logic and evolutionary systems).

Large margin classifiers (such as MLPs) are designed to assign training samples with high confidence (or margin) to one of the classes. Recent theoretical results of these systems show why the use of regularisation terms and feature extractor techniques can enhance their generalisation properties. Since the optimal subset of features selected depends on the classification problem, but also on the particular classifier with which they are used, global learning algorithms for large margin classifiers that use feature extractor techniques are desired. A direct approach is to optimise a cost function based on the margin error, which also incorporates regularisation terms for controlling capacity. These terms must penalise a classifier with the largest margin for the problem at hand. Our work shows that the inclusion of a PCA term can be employed for this purpose. Since PCA only achieves an optimal discriminatory projection for some particular distribution of data, the margin of the classifier can then be effectively controlled. We also propose a simple constrained search for the global algorithm in which the feature extractor and the classifier are trained separately. This allows a degree of flexibility for including heuristics that can enhance the search and the performance of the computed solution. Experimental results demonstrate the potential of the proposed method.

This letter addresses the asymptotic convergence of Kohonen's LVQ1 algorithm when the number of training samples are finite with an analysis that uses the dynamical systems and optimisation theories. It establishes the sufficient conditions to ensure the convergence of LVQ1 near a minimum of its cost function for constant step sizes and cyclic sampling. It also proposes a batch version of LVQ1 based on the very fast Newton optimisation method that cancels the dependence of the on-line version on the order of supplied training samples.

This paper introduces a learning strategy for designing a set of prototypes for a 1-nearest-neighbour (NN) classifier. In learning phase, we transform the 1-NN classifier into a maximum classifier whose discriminant functions use the nearest models of a mixture. Then the computation of the set of prototypes is viewed as a problem of estimating the centres of a mixture model. However, instead of computing these centres using standard procedures like the EM algorithm, we derive to compute a learning algorithm based on minimising the misclassification accuracy of the 1-NN classifier on the training set. One possible implementation of the learning algorithm is presented. It is based on the online gradient descent method and the use of radial gaussian kernels for the models of the mixture. Experimental results using hand-written NIST databases show the superiority of the proposed method over Kohonen's LVQ algorithms.

A novel classifier is introduced to overcome the limitations of the k-NN classification systems. It estimates the posterior class probabilities using a local Parzen window estimation with the k-nearest-neighbour prototypes (in the Euclidean sense) to the pattern to classify. A learning algorithm is also presented to reduce the number of data points to store. Experimental results in two hand-written classification problems demonstrate the potential of the proposed classification system.

We introduce a batch learning algorithm to design the set of prototypes of 1 nearest-neighbour classifiers. Like Kohonen's LVQ algorithms, this procedure tends to perform vector quantization over a probability density function that has zero points at Bayes borders. Although it differs significantly from their online counterparts since: (1) its statistical goal is clearer and better defined; and (2) it converges superlinearly due to its use of the very fast Newton's optimization method. Experiments results using artificial data confirm faster training time and better classification performance than Kohonen's LVQ algorithms.

This paper evaluates a set of computational algorithms for the automatic estimation of human postures and gait properties from signals provided by an inertial body sensor. The use of a single sensor device imposes limitations for the automatic estimation of relevant properties, like step length and gait velocity, as well as for the detection of standard postures like sitting or standing. Moreover, the exact location and orientation of the sensor is also a common restriction that is relaxed in this study. Based on accelerations provided by a sensor, known as the `9 2', three approaches are presented extracting kinematic information from the user motion and posture. Firstly, a two-phases procedure implementing feature extraction and Support Vector Machine based classi cation for daily living activity monitoring is presented. Secondly, Support Vector Regression is applied on heuristically extracted features for the automatic computation of spatiotemporal properties during gait. Finally, sensor information is interpreted as an observation of a particular trajectory of the human gait dynamical system, from which a reconstruction space is obtained, and then transformed using standard principal components analysis, nally Support Vector Regression is used for prediction. Daily living Activities are detected and spatiotemporal parameters of human gait are estimated using methods sharing a common structure based on feature extraction and kernel methods. The approaches presented are susceptible to be used for medical purposes.

This paper presents an algorithm for the automatic estimation of spatio temporal gait properties from signals provided by inertial body sensors. The approach is based on time series analysis. Here, a minimum number of body sensor devices is used, which imposes limitations for the automatic extraction of relevant properties of the gait like step length and velocity. The human gait is represented as a dynamical system (DS), which internal states are hidden. Sensor information is interpreted as an observation of a particular trajectory of the DS, from wich a reconstruction space can be obtained. The reconstruction space is then transformed using standard principal components analysis (PCA). From the transformed space, reliable models to estimate step length and velocities are successfully constructed.

User daily activity monitoring is useful for physicians in geriatrics and rehabilitation as a indicator of user health and mobility. Real time activities recognition by means of a processing node including a triaxial accelerometer sensor situated in the user's chest is the main goal for the presented experimental work. A two-phases procedure implementing features extraction from the raw signal and SVM-based classification has been designed for real time monitoring. The designed procedure showed an overall accuracy of 92% when recogninzing experimentation performed in daily conditions.

10 pages, 7 figures, 4 tables, The present study reviewed some of the critical pre-processing steps required for otolith shape characterisation for automatic classification with heterogeneous distributed data. A common procedure for optimising automatic classification is to apply data pre-processing in order to reduce the dimension of vector inputs. One of the key aspects of these pre-processing methods is the type of codification method used for describing the otolith contour. Two types of codification methods (Cartesian and Polar) were evaluated, and the limitations (loss of information) and the benefits (invariance to affine transformations) associated with each method were pointed out. The comparative study was developed using four types of shape descriptors (morphological, statistical, spectral and multiscale), and focused on data codification techniques and their effects on extracting shape features for automatic classification. A new method derived from the Karhunen–Loève transformation was proposed as the main procedure for standardising the codification of the otolith contours., The present study was funded by the IBACS project (European Union Project QLRT-2001-01610), by the Spanish project AVG-ION (McyT-TIC2000-0376-p4-04) from the Spanish Ministry of Science and Technology and by the IBACS European project.

In this paper an attempt to design decision part of the iris identification system, which will be able to identify persons just by a look at the camera, is discussed. Proposed system will be cheap in comparison with present ones. Modified Haralick's co-occurrence method with multilayer perceptron is used for extraction and classification of the irises. Some preliminary research results are also presented.

Ion-selective field transistors (ISFET) are electronic devices that merge solid-state electronic technology with chemical sensors for being sensitive to the concentration of a particular ion in a solution. However, as it has been reported, their response does not only depend on a single ion but is also affected by several interfering ions found in the solution to be measured. These interfering ions can be considered as a noise and consequently, a post-processing stage that increases the SNR is mandatory. Our work shows how neural network (NN) structures, a kind of statistical learning machines, could be used for this purpose. In particular, we introduce several novel neural learning architectures for ISFET source separation from interfering ions, which employ ISFET models as a prior knowledge. The proposed NNs are grouped in two categories: supervised and unsupervised. The supervised NN is a RBF-like solution that could be used using a single ISFET or with an ISFET array. On the other hand, the unsupervised NN is based on non-linear independent component analysis (ICA) that employs two or more ISFETs. Since the RBF-like NN structure needs many training data pairs for calibration, and this could be a practical problem, a synergistic combination of unsupervised and supervised methods is introduced. The proposed hybrid NN is based on a combination of non-linear ICA and a linear predictor that considerably reduces the number of required samples in order to achieve a good solution. In the final work, several experimental results are included, which demonstrate the interest and viability of the proposed solution. The work is in progress, as a part of the SEWING EU project (contract IST-2000-28084).

Large margin classifiers are computed to assign patterns to a class with high confidence. This strategy helps controlling the capacity of the learning device so good generalization is presumably achieved. Two recent examples of large margin classifiers are support vector learning machines (SVM) [12] and boosting classifiers[10]. In this paper we show that it is possible to compute large-margin maximum classifiers using a gradient-based learning based on a cost function directly connected with their average margin. We also prove that the use of this procedure in nearestneighbor (NN) classifiers induce solutions closely related to support vectors.

In this chapter we address the applicability of artificial neural network and fuzzy logic models to real tasks in industrial environments. For this purpose, we shall present a general methodology which will be outlined by means of a case study which consists in the implementation of a classification/decision engine included in an automatic coin recogniser. This coin recogniser is a part of currently available commercial vending machines. The methodology presented can be considered as divided in three main tasks: database compilation, selection of the proper neural or fuzzy model and implementation. A wide range of models, including classical as well as evolutionary algorithms, has been considered. The experimental results demonstrate that the use of artificial neural and fuzzy models overcomes some of the limitations inherent in the traditional techniques considered when solving this task.

We present two online gradient learning algorithms to design condensed k-nearest neighbor (NN) classifiers. The goal of these learning procedures is to minimize a measure of performance closely related to the expected misclassification rate of the k-NN classifier. One possible implementation of the algorithm is given. Converge properties are analyzed and connections with other works are established. We compare these learning procedures with Kononen’s LVQ algorithms [7] and k-NN classification using the handwritten NIST databases [5]. Experimental results demonstrate the potential of the proposed learning algorithms.

In this paper we shall present a general methodology to be used when artificial neural network models are applied to solve real tasks in industrial environments. This methodology will be outlined by means of a case study which consists in the implementation of the decision/classification engine to be included in an automatic coin classifier. This coin classifier is incorporated in commercial vending machines, so that the problems arising when trying to face the conditions imposed by real environments have to be considered. The methodology presented in this paper can be considered as divided in three main tasks: database compilation and analysis, selection of the proper neural model and its implementation. A wide range of neural models, including classical as well as evolutive algorithms, has been considered. As the experimental results provided show, the use of artificial neural models for implementing the proposed classifier proves to overcome some of the limitations inherent to the traditional techniques considered when solving this task.

Two different approaches in constructing Neural Network (NN) classifiers are discussed — discriminant-based networks and Region of Influence networks. A general model for ROI networks is presented, and the different functionalities of this structure are discussed: classification, vector quantization and associative memory.

10 pages, 9 figures, 3 tables, Fish otolith morphology has been closely related to landmark selection in order to establish the most discriminating points that can help to differentiate or find common characteristics in sets of otolith images. Fourier analysis has traditionally been used to represent otolith images, since it can reconstruct a version of the contour that is close to the original by choosing a reduced set of harmonic terms. However, it is difficult to locate the contour’s singularities from this spectrum. As an alternative, wavelet transform and curvature scale space representation allow us to quantify the irregularities of the contour and determine its precise position. These properties make these techniques suitable for pattern recognition purposes, ageing, stock determination and species identification studies. In the present study both techniques are applied and used in an otolith classification system that shows robustness against affine image transformations, shears and the presence of noise. The results are interpreted and discussed in relation to traditional morphology studies., The current work was supported by the Spanish project MYCYT TIC2000-0376-P4-04.

In order to prevent and reduce water pollution, promote a sustainable use, protect the environment and enhance the status of aquatic ecosystems, this article deals with the application of advanced mathematical techniques designed to aid in the management of records of different water quality monitoring networks. These studies include the development of a software tool for decision support, based on the application of fuzzy logic techniques, which can indicate water quality episodes from the behaviour of variables measured at continuous automatic water control networks. Using a few physicalchemical variables recorded continuously, the expert system is able to obtain water quality phenomena indicators, which can be associated, with a high probability of cause-effect relationship, with human pressure on the water environment, such as urban discharges or diffuse agricultural pollution. In this sense, at the proposed expert system, automatic water quality control networks complement manual sampling of official administrative networks and laboratory analysis, providing information related to specific events (discharges) or continuous processes (eutrophication, fish risk) which can hardly be detected by discrete sampling.

Otoliths are calcified structures in the inner ear of fish. The otolith shape changes during a fish's lifetime are particular to individual species. Then, otolith shape can be used to differentiate between species and between fish of the same species. Fishery research has used the growth patterns (i.e. rings) found in these calcified structures to estimate the age of individual fish. However, many factors, such as seasonal variations, temperature, habitat and food, may influence otolith growth. Then, the manual classification of otoliths remains a difficult task, and even experienced examiners can give inaccurate age estimation. We propose to use statistical learning techniques (artificial neural networks) to improve and automate the process. ANN classification methods are evaluated and used with some real otolith databases, giving significant results

In order to extract learning algorithms from living neural aggregates it would be advantageous to achieve one-to-one neuron-electrode interfacing with in vitro networks. Towards this goal, we have developed a hybrid glass-elastomer technology, which allows topology specification in small networks (of the order of 10 neurons) and recording of extracellular potentials from individual neurites grown through microfluidic channels. Here we report on progress towards adhesion-free placement of cells within microwells, promotion of neurite growth and recording of intra-channel extracellular spikes.