Artificial Computation in Biology and Medicine

Computer aided diagnosis (CAD) constitutes an important tool for the early diagnosis of Alzheimer’s Disease (AD), which, in turn, allows the application of treatments that can be simpler and more likely to be effective. This paper presents a straightfoward approach to determine the most discrimanative brain regions, defined by the Automated Anatomical Labelling (AAL), based on the measurements of the tissue density at the different brain areas. Statistical analysis of GM and WM densities reveal significant differences between controls (CN) and AD at specific brain areas associated to tissue density diminishing due to neurodegeneration. The proposed method has been evaluated using a large dataset from the Alzheimer’s disease Neuroimaging Initiative (ADNI). Classification results assessed by cross-validation proved that computed WM/GM densities are discriminative enough to differentiate between CN/AD. Moreover, fusing density measurements with ApoE genetic information help to increase the diagnosis accuracy.

In the present paper we describe an artificial neural model of the Number Interval Position Effect (NIPE;[5]) that has been observed in the mental bisection of number intervals both in adults and in children. In this task a systematic error bias in the mental setting of the subjective midpoint of number intervals is found, so that for intervals of equal size there is a shift of the subjective midpoint towards numbers higher than the true midpoint for intervals at the beginning of decades while for intervals at the end of decades the error bias is directionally reversed towards numbers lower than the true midpoint. This trend of the bisection error is recursively present across consecutive decades.

Here we show that a neural-computational model based on information spread by energy gradients towards accumulation points based on the logarithimic compressed representation of number magnitudes that has been observed at the single cell level in rhesus monkeys [9] effectively simulates the performance of adults and children in the mental bisection of number intervals, in particular replicating the data observed in children.

Alzheimer’s Disease (AD) is nowadays the most common type of dementia, with more than 35.6 million people affected, and 7.7 million new cases every year. Magnetic Resonance Imaging (MRI) is a fairly widespread tool used in clinical practice, and has repeatedly proven its utility in the diagnosis of AD. Therefore a number of automatic methods have been developed for the processing of MR images. In this work, a new algorithm that projects the three-dimensional image onto two-dimensional maps using Local Binary Patterns (LBP) is presented. The algorithm yields visually-assessable maps that contain the textural information and achieves up to a 90.5% accuracy in a differential diagnosis task (AD vs controls), which proves that the textural information retrieved by our methodology is significantly linked to the disease.

This paper presents the design, development, simulation and test of a directional telemetry system for cochlear implants using FPGA. We used Manchester codification and ASK modulation in order to achieve a high transmission speed. The design was simulated using the System Generator for FPGA by Xilinx and Simulink developed by Mathworks. Also, the design was emulated using the ISE design software by Xilinx. The design has been tested under noisy environment. The design was optimised so as to obtain a power consumption equal or less than the maximum allowed in the receiver. We achieved the use fewer components of the FPGA. As a result, the telemetry system has been designed to meet with specifications for use it in the development of a prototype of cochlear implant for research purposes.

In this work we examine the problem of finding biological motifs in DNA databases. The problem was solved by applying MBMEDA, which is a evolutionary method based on the Estimation of Distribution Algorithm (EDA). Though it assumes statistical independence between the main variables of the problem, results were quite satisfactory when compared with those obtained by other methods; in some cases even better. Its performance was measured by using two metrics: precision and recall, both taken from the field of information retrieval. The comparison involved searching a motif on two types of DNA datasets: synthetic and real. On a set a five real databases the average values of precision and recall were 0.866 and 0.798, respectively.

The retina is one of the most extensively studied neural circuits in the Visual System. Numerous models have been proposed to predict its neural behavior on the response to artificial and natural visual patterns. These models can be considered an important tool for understanding the underlying biophysical and anatomical mechanisms. This paper describes a general-purpose simulation environment that fits to different retina models and provides a set of elementary simulation modules at multiple abstraction levels. The platform can simulate many of the biological mechanisms found in retinal cells, such as signal gathering though chemical synapses and gap junctions, variations in the receptive field size with eccentricity, membrane integration by linear and single-compartment models and short-term synaptic plasticity. A built-in interface with neural network simulators reproduces the spiking output of some specific cells, such as ganglion cells, and allows integration of the platform with models of higher visual areas. We used this software to implement whole retina models, from photoreceptors up to ganglion cells, that reproduce contrast adaptation and color opponency mechanisms in the retina. These models were fitted to published electro-physiological data to show the potential of this tool to generalize and adapt itself to a wide range of retina models.

In the olfactory system we can observe two types of neurons based on their responses to odorants. Specialist neurons react to a few odorants, while generalist neurons respond to a wide range of them. These kinds of neurons can be observed in different parts of the olfactory system. In the antennal lobe (AL), these neurons encode odorant information and in the extrinsic neurons (ENs) of the mushroom bodies (MB) they can learn and identify different kind of odorants based on the selective and generalist response. The classification of specialists and generalists neurons in Kenyon cells (KCs), which serve as a bridge between AL and ENs, may seem arbitrary. However KCs have the unique mission of increasing the separability between different odorants, to achieve a better information processing performance. To carry out this function, the connections between the antennal lobe and Kenyon cells do not require a specific connectivity pattern. Since KCs can be specialists or generalists by chance and olfactory learning performance relies on their feature extraction capabilities, we analyze the role of generalist and specialist neurons in an olfactory discrimination task. Role that we studied by varying the percentage of these two kind of neurons in KC layer. We determined that specialist neurons are a decisive factor to perform optimal odorant classification.

The intensity normalization step is essential, as it corresponds to the initial step in any subsequent computer-based analysis. In this work, a proposed intensity normalization approach based on a predictive modeling using multivariate linear regression (MLR) is presented. Different intensity normalization parameters derived from this model will be used in a linear procedure to perform the intensity normalization of

123

I-ioflupane-SPECT brain images. This proposed approach is compared to conventional intensity normalization methods, such as specific-to-non-specific binding ratio, integral-based intensity normalization and intensity normalization by minimizing the Kullback-Leibler divergence. For the performance evaluation, a statistical analysis is used by applying the Euclidean distance and the Jeffreys divergence. In addition, a classification task using support vector machine to evaluate the impact of the proposed methodology for the development of a computer aided diagnosis (CAD) system for Parkinsonian syndrome detection.

An accurate and early diagnosis of the Alzheimer’s disease (AD) is of fundamental importance to improve diagnosis techniques, to better understand this neurodegenerative process and to develop effective treatments. In this work, a novel classification method based on independent component analysis (ICA) and supervised learning methods is proposed to be applied on segmented brain magnetic resonance imaging (MRI) from Alzheimer’s disease neuroimaging initiative (ADNI) participants for automatic classification task. The ICA-based method is composed of three step. First, MRI are normalized and segmented by the Statistical Parametric Mapping (SPM8) software. After that, average image of normal (NC), mild cognitive impairment (MCI) or AD subjects are computed. Then, FastICA is applied to these different average images for extracting a set of independent components (IC) which symbolized each class characteristics. Finally, each brain image from the database was projected onto the space spanned by this independent components basis for feature extraction, a support vector machine (SVM) is used to manage the classification task. A 87.5% accuracy in identifying AD from NC, with 90.4% specificity and 84.6% sensitivity is obtained. According to the experimental results, we can see that this proposed method can successfully differentiate AD, MCI and NC subjects. So, it is suitable for automatic classification of sMRI images.

With regard to neural networks, there are two different areas which have generated two lines of research. One research interest comes from the field of computer science which seeks to create and design neural networks capable of performing computational tasks. In this line of research, any neural network is relevant because the important issue is the problems which they are capable of resolving. Thus, neural networks are computational devices and computational power and the computational process which they perform are researched. The other interest of research is related to neuroscience. This focuses on both neural and brain activity. The big difference between these two lines of research can be observed from the outset. In the first, the neural network is designed and its performance on computational tasks is then researched. In the second, performance on computational tasks is known but the neural mechanism is not and neuroscience seeks to identify it. An interaction between these two lines of research is very positive because it produces synergies which generate important advances in both lines of research e.g. Hopfield’s networks. This article enunciates a neural mechanism to interpret neural dynamics based on some of the results produced by computer science. This mechanism identifies an internal or external state

s

with a formal language

L

. Independently, if the mechanism exist or not in the human brain, this mechanism can be used to design new architectures for neural networks.

The use of image analysis in understanding how powdered coal burns during the combustion plays a significant role in setting combustion parameters. During the pulverised coal combustion, char particles are produced by devolatising coal and represent the dominant stage in the combustion process. The pyrolysis produces different char morphologies that determine coal reactivity affecting the performance of coal combustion in power plants and the emissions of carbon dioxide, CO2. In this paper, an automatic char classification model is proposed using supervised learning. A general classification model is trained given a set of char particles classified by an expert. In particular, Support Vector Machine (SVM) and Random Forest are the trained classifiers. Two types of features are evaluated to built classification models: local and global. Local features are calculated using the Scale-Invariant Transform Feature (SIFT). Global features are defined based on the morphology classification by the International Committee for Coal and Organic Petrology (ICCP). Each classifier is trained by SVM or Random Forest and evaluated using a 10-fold cross-validation. The 70% of data is used as training set and the rest as testing set. A total of 2928 char-particle images are used for evaluating performance of classification models. Additionally, evaluation of model generalisation capability is done using a test set of 732 char particle images. Results showed that global features – defined by the application domain – increase significantly the accuracy of classifiers. Also, global features have more generalisation power than local features. Local features lack of meaning in the application domain and classifiers build with local features – such as SIFT – depend crucially on the training set.

The retina is responsible for transducing visual information into spikes trains which are then sent via the optical nerve to the visual cortex. This is the first step in the visual pathway responsible for the sense of vision. Our research group is working on the design of a cortical visual neuroprosthesis aimed to restore some functional vision to profoundly visual-impaired people. The goal of developing such a bioinspired retinal encoder is not simply to record a high-resolution image, but to process its visual information and transmit it in a meaningful way to the appropriate area on the visual cortex. Retinal models to be implemented have to match as much as possible the output produced by an actual biological retina. The models involve a big search space defined by a set of parameters that have to be appropriately adjusted. This in itself has several problems which need to be addressed. We propose in this paper an automatic evolutionary multi-objective strategy for selecting those parameters which best approximate the outputs by the synthetic retina model and the biological records. A case study is presented where results of a retina model tuned with our method are compared to biological recordings.

This study investigates the effect of two personality traits, dominance and extroversion, on social intelligence. To test these traits, a NAO robot was used, which was teleoperated through a computer using a Wizard of Oz technique. A within-subject design was conducted with extroversion as within-subject variable and dominance as between-subject. Participants were asked to cooperate with the robot to play “Who wants to be a millionaire”. Before the experiment participants filled in a personality questionnaire to measure their dominance and extroversion. After each condition, participants filled in a modified version of the Godspeed questionnaire concerning personality traits of the robot plus 4 extra traits related to social intelligence. The results reveal a significant effect of dominance and extroversion on social intelligence. The extrovert robot was judged as more socially intelligent, likeable, animate, intelligent and emotionally expressive than the introvert robot. Similarly, the submissive robot was characterized as more socially intelligent, likeable and emotionally expressive than the dominant robot. We found no substantial results towards the similarity-attraction hypothesis and therefore we could not make a conclusion about the mediating effect of participant” s personality on likeability.

Artificial Metaplasticity are Artificial Learning Algorithms based on modelling higher level properties of biological plasticity: the plasticity of plasticity itself, so called Biological Metaplasticity. Artificial Metaplasticity aims to obtain general improvements in Machine Learning based on the experts generally accepted hypothesis that the Metaplasticity of neurons in Biological Brains is of high relevance in Biological Learning. Artificial Metaplasticity Multilayer Perceptron (AMMLP) is the application of Metaplasticity in MLPs ANNs trying to improve uniform plasticity of the Backpropagation algorithm. In this paper two different AMMLP algorithms are applied to the MIT-BIH electro cardiograms database and results are compared in terms of network performance and error evolution.

In the last years there has been a noticeable progress in motor learning, neuroplasticity and functional recovery after the occurrence of brain lesion. Rehabilitation of motor function has been associated to motor learning that occurs during repetitive, frequent and intensive training. Neuro-rehabilitation is based on the assumption that motor learning principles can be applied to motor recovery after injury, and that training can lead to permanent improvements of motor functions in patients with muscle deficits. The emergent research field of Rehabilitation Engineering may provide promise technologies for neuro-rehabilitation therapies, exploiting the motor learning and neural plasticity concepts. Among those technologies, the FES-assisted systems could provide repetitive training-based therapies and have been developed to aid or control the upper and lower limbs movements in response to user’s intentionality. Surface electromyography (SEMG) reflects directly the human motion intention, so it can be used as input information to control an active FES-assisted system. The present work describes a neurorehabilitation platform at the upper-limb level, based on bilateral coordination training (i.e. mirror movements with the unaffected arm) using a close-loop active FES system controlled by user. In this way, this work presents a novel myoelectric controller for decoding movements of user to be employed in a neurorehabilitation platform. It was carried out a set of experiments to validate the myoelectric controller in classification of seven human upper-limb movements, obtaining an average classification error of 4.3%. The results suggest that the proposed myoelectric pattern recognition method may be applied to control close-loop FES system.

This work presents a study that evaluates different scenarios of preprocessing and processing of EEG registers, with the aim to predict fist imaginary movements utilizing the data of the EEG Motor Movement/Imaginary Dataset. Three types of imaginary fist movements have been decoded: sustained opening and closing of right fist, sustained opening and closing of left fist and rest. Initially, the registers were band-pass filtered to separate frequency ranges given by

mu

rhythms (7.5-12.5 Hz),

beta

rhythms (12.5-30 Hz),

mu

&

beta

rhythms, and a

broad

range of 0.5-30 Hz. Afterward, the signals of the separated subbands were epoched in time windows of 0-0.5, 0-1, 0-1.5 and 0-2 seconds, as well as preprocessed with two techniques of spatial filtering: common spatial patterns and independent component analysis. In both cases, a set of selected channels was established for feature extraction, by calculation of the logarithms of the variance in the time series corresponding to each preprocessed and selected channel. The classification stage was based on linear discriminant analysis and support vector machines. The results showed that the combination given by common spatial patterns and support vector machines allowed to reach a mean decoding accuracy close to 99.9%, where epoching and filtering to separate subbands did not influence the results in a noticeable way.

In this paper we present a new unsupervised neural network whose architecture resembles the koniocortex, the first cortical layer receiving sensory inputs. For easiness, its properties were incorporated in a step by step manner along successive network versions. In some cases, the version improvement consists in the replacement of a non-biological property by a biologically plausible one. Initially (version 0) the network was merely an scaffold implementing the Bayes Decision Rule. The first network version incorporated metaplasticity and intrinsic plasticity, but neural competition was not biological. In a second version, competition naturally occurred due to the interplay between lateral inhibition and homeostatic properties. Finally, in the koniocortex-like network, competition and pattern classification emerges naturally due to the interplay of inhibitory interneurons and previous version’s properties. An example of numerical character recognition is presented for illustrating the main characteristics of the network.

Medical institutions are increasingly aware of the vast amount of available data they have and its potential benefits. These data are being analyzed and shared at institutions all around the world, however, the way the data are stored, managed and secured need for new technological solutions to facilitate its consumption and sharing between institutions. This situation has become a technological challenge for the interoperability, data mining and Big Data fields. Neuroimaging community is one of the most active in looking for effective solutions, like the XNAT project which aims for neuroimaging data acquisition, management and processing. This paper shows the ongoing effort to develop a Semantic Framework to facilitate multidimensional data analysis based on XNAT architecture.

Hallucinations, and more specifically auditory hallucinations (AH), are a perplexing phenomena experienced by many people. Though they are a clinical symptom in some mental diseases, such as Schizophrenia, they are also experienced by normal, healthy persons. There are several models of the mechanics happening in the brain leading to hallucinations, which involve auditory, language and emotion regions. On the other hand, there is not much empirical evidence due to the evanescence of the phenomena, and the difficulty to capture meaningful data. Recent works on resting state functional Magnetic Resonance Imaging (rs-fMRI) data, are providing confirmation of some brain localizations. Dynamic Causal Modeling (DCM) provides estimations of neural effective connectivity parameters from the experimental fMRI data, and recently has been proposed to work on rs-fMRI data. We provide preliminar results on a dataset that recently has been useful to find confirmation of AH model effects.

This paper analyzes the filtering operation carried out by the classical Difference-of-Gaussians model proposed by Rodieck to describe the receptive fields of retinal ganglion cells. Discrete DoG kernels of such functions were developed and compared with High-Pass and High-Frequency Enhancing filters. The results suggest that the DoG Kernels behave as High-Frequency Enhancing filters but in a limited band of frequencies.

At present there is a growing interest in studying emotions in the brain. However, although in the latest years there have been numerous studies, little is known about their temporal dynamics. Techniques such as fMRI or PET have very good spatial resolution but poor temporal resolution and vice-versa in the case of EEG. In this study we propose to use EEG to gain insight into the spatiotemporal dynamics of emotions processing with a better time resolution. We conducted an experiment in which binary classification (like / dislike) of standardized images was performed. Topographic changes in EEG activity were examined in the time domain. In the spatial dimension, we used a rotating dipole for the spatial location and determination of Cartesian coordinates (x, y and z). Our results showed a temporal window (424-474msec) with a significant difference which involved a lateralization (left to very positive stimuli and right to very negative stimuli) even for neutral stimuli. These results support the lateralization of brain activity during processing of emotions.

Understanding the underlying properties of neuronal populations over single neurons is a longstanding goal for both basic and applied neurosciences, with a specifically suitable application in the field of neuroprosthesis development, aimed to restore the loss of function of a visual cortex as a result of an injury or disease. We study how the interstimulus interval (ISI) period of a repeated visual stimulus influences the overall activity of rat visual cortex neuronal populations. Our results suggest that certain (3, 5 s) interstimulus intervals do have an increased stimulus response compared to longer or shorter ISIs for a 500 ms grating drifting stimulus. Based on the preliminary results shown in this article, we claim the need of a better understanding of the biological dynamics of the visual cortex neuronal populations in order to properly design suitable brain-machine interfaces for visual neurorehabilitation intracortical neuroprosthetics.

One of the many important functions the brain carries out is interpreting the external world. For this, one sense that most mammals rely on is vision. The first stage of the visual system is the image processing whose capture takes place in the retina. Here, photoreceptors cells transform light into electrical impulses that are then guided by amacrine, bipolar, horizontal and some glial cells up to the ganglion cells layer. Ganglion cells decode the visual information to be interpreted by the visual cortex. The understanding of the mechanism for decoding the visual information is a major task and challenge in neuroscience. This is especially true for images that change with time, for example during movement. For this purpose, extracellular recordings with a 100 multi-electrode-array (MEA) were carried out in the retinal ganglion cells layer of mice. Different moving patterns and actual images were used to stimulate the retina. Here, we present a new strategy for analysis over the spike trains recorded allowing the reconstruction of the actual stimuli with a reduced number of ganglion cell responses.

Electric stimulation in neural cultures in neural cultures may be used for creating adjacent physical or logical connections in the connectivity graph following Hebb’s Law modifying the neural responses principal parameters. The created biological structure may be used for computing a certain function, however this achieved structure vanished with time as the stimulation stops. A DTCNN architecture, specifically designed for optimum parallel implementation over dedicated hardware, is proposed to emulate the behavior ans structure of the biological neuronal culture. The FPGA circuit can be used as a permanent model and is also intended to facilitate and speed up further experimentation.

Organic as well as neurologic diseases leave important correlates in phonation. Parkinson’s Disease (PD) may leave marks in vocal fold dystonia and tremor. Biomechanical parameters monitoring vocal fold tension and unbalance, as well as tremor are defined in the study. These correlates are known to be of help in tracing the neuromotor activity of both laryngeal and articulatory pathways. As the population affected by PD is mainly above 60, the main problem found is how to differentiate PD phonation correlates from aging voice (presbyphonia). An important objective is to explore which correlates react differentially to PD than to aging voice. As an example a study is conducted on a set of male PD patients being monitored in short intervals by recording their phonation. The results of these longitudinal studies are presented and discussed.

This paper aims at analyzing the effects of the discretization process of the continuous Difference-of-Gaussians models obtained empirically by Enroth-Cugell and Robson (1966). The filter properties of the Discrete DoG kernels were analyzed in the frequency domain and their effects on input images were characterized by means of GLCM descriptors. The results demonstrate that the DoG Kernels behaviour range between true High-Frequency Enhancing filters and Band-pass filters depending on the discretization parameters. Moreover, the analysis of filtered images suggest that those kernels that enhance contrast come at a cost of higher entropy as well as lower spatial correlation.

The overuse of antimicrobials promotes the resistance of antibiotics, which is a great concern in hospitals. Clinical Guidelines are essential documents that provide useful recommendations to clinicians about the therapy. In order to obtain a Computerised Clinical Guideline, main efforts to represent this knowledge focus on ad-hoc data flow models. However, they have had a low impact in the industry since they generally neglect clinical standards or they are hard to maintain due to the model complexity. In this work, we propose to step backward to use rule-based approaches to obtain clinical rules, more simple to model and easier to manage. We also review and discuss main rule representation alternatives and we present a case study in the Ventilator Associated Pneumonia from a Clinical Guideline.

Many diagnosis support systems (DSS) are focused on precise disorders, being not useful for differential diagnosis (DD) or facing comorbidities. Few DSSs offer a rich list of potential diagnoses and they do not reflect complex relations between diseases to be diagnosed. We present a model to allow collaboration of multiple heterogeneous diagnostic units (DU), which are actual DSSs, behaving as a whole system. The heterogeneity of the DUs refers to the disease they diagnose and the classification model they use to do so. This model offers a framework to build multi-purpose DSSs, assuring their operability and functioning despite the heterogeneity of the single diagnostic units.

The ability of walking brings us a great freedom in our daily life. However, there is a huge number of people who have this ability diminished or are not even able to walk due to motor disabilities. This paper presents a method to detect the voluntary initiation and stop of the gait cycle using the ERD phenomenon. The system developed obtains a good accuracy in the detection of the rest and walking state (70.5 % and 75.0 %, respectively). Moreover, the average detection of the onset and ending instants of the gait is detected with a 65.2 % of accuracy. Taking into account the number of intentions of initiation and stop of the gait, the system reaches a good True Positive Rate (around 65%) but obtaining a still improvable False Positive Rate (15.4 FP/min in average). By reducing this factor, this detection system can be used in future works to control a lower limb exoskeleton or a wearable robot. These devices are very useful for rehabilitation and assistance procedures in patients with motor problems affecting their lower limb.

The great usefulness of remote recording of biomedical signals in most aspects of daily life has generated an increasing interest in this field. Traditionally, monitoring devices from clinical enviroments are bulky, intrusive, and expensive. Thus, the development of wearable, mobile, and low-cost applications is desirable. Nevertheless, recent improvements in open-hardware allow developing low cost devices and portable designs for biosignal monitoring in out-of-lab applications, such as sports, leisure, e-Health, etc. This paper presents a low-cost wearable system able to simultaneously record electrical brain and heart activity (i.e. electroencephalography and electrocardiography). The system is able to send biomedical data to a platform for remote analyses. Both software and hardware are open-source. We assessed the system for its validity and reliability in a real road environment.

The aging issue threatens to collapse health public systems in some regions of first world. Although telemedicine is one of the solutions to avoid people insti-tutionalization, it has severe limitations and not all medical services can be of-fered. While few years ago the electrical complexity and cost of EEG systems prevented execution of clinical EEG tests out of hospital, now services such as home-based video-EEG are possible. Conversely, some important clinical tests such as event-related potentials cannot be executed remotely. The reason for that is the accurate synchrony between local stimulus onset and remote starting of EEG acquisition. In hospital, synchrony is guaranteed by means of a wired connection between stimulus display that triggers EEG recording while in home-based testing this link normally does not exist. In this study we show an effective way to execute event-related potentials based on asynchronous EEG data transmission. We executed a dichotic listening paradigm with forced-attention modality. The user goal was to detect the attended audio sentence from the analysis of evoked auditory event-related potentials. The rate of successful detection in both synchronous and asynchronous modalities was compared and results revealed no significant difference. Our asynchronous approach can be used in on-line acquisition of home-based event-related potentials with remote processing.

Chronic Obstructive Pulmonary Disease (COPD) places an enormous burden on the health care systems and causes diminished health related quality of life. The highest proportion of human and economic cost is associated to admissions for acute exacerbation of respiratory symptoms. The remote monitoring of COPD patients with the view of early detection of acute exacerbation of COPD (AECOPD) is one of the goals of the respiratory community. In this study, machine learning was used to develop predictive models. Models robustness to exacerbation definition was analyzed. A non-knowled-ge based approach was followed on data self-reported by patients using a multimodal tool during a remote monitoring 6 months trial. Comparison of different classifier algorithms operating with different AECOPD definitions was performed. Significant results were obtained for AECOPD prediction, regardless of the definition of exacerbation used. Best accuracy was achieved using a PNN classifier independently of the selected AECOPD definition. Our study suggests that the proposed data-driven methodology could help to design reliable predictive algorithms aimed to predict COPD exacerbations and therefore could provide support both to physicians and patients.

We explore a novel approach in which BCI input is used to influence the behaviour of search algorithms which are at the heart of many Intelligent Systems. We describe how users can influence the behaviour of heuristic search algorithms using Neurofeedback (NF), establishing a connection between their mental disposition and the performance of the search process. More specifically, we used functional near-infrared spectroscopy (fNIRS) to measure frontal asymmetry as a marker of approach and risk acceptance under a NF paradigm, in which users increased their left asymmetry. Their input was mapped onto a dynamic weighting im- plementation of A* (termed WA*), modifying the behaviour of the algorithm during the resolution of an 8-puzzle problem by adjusting the performance-optimality tradeoff. We tested this approach with a proof-of-concept experiment involving 11 subjects who had been previously trained in NF. Subjects were able to positively influence the behaviour of the search process in over 58% of the NF epochs, resulting in faster solutions.

This work proposes a procedure to measure the human capability to discriminate couples of English vocalic phonemes embedded into words. Using the analysis of the EEG response to auditory contrasts in an oddball paradigm experiment, the Medium Mismatch Negativity potential (

MMN

) is evaluated. When the discrimination is achieved, MMN has a negative amplitude while positive or zero MMN amplitudes correspond to the confusion of the two vocalic phonemes heard by the subject performing the experiment. The procedure presented has many potential usages for phonetic learning tools given its capability to automatically analyze discrimination of sounds. This permits its usage in interactive and adaptive applications able to keep track of the improvements made by the users.

In the past few years, numerous efforts have been devoted to the segmentation and characterization of the human heart from various medical image techniques. This paper addresses the first results of parametric deformable models defined in the Fourier domain as a tool to characterize the shape of the heart. The main advantage of these models is their high speed of adaptation to the dataset and their robustness against noise. In addition, due to their explicit parametric typology, different parameters of its dynamical behaviour can be derived from their mathematical expression. This article details the mathematical framework of deformable models defined in the frequency domain as well as the preprocessing and practical implementation of the model used in this application to model the cardiac cycle of the whole heart.

This paper describes an application of variational image registration. The method is based on an efficient implementation of the diffusion registration formulated in the frequency domain. The goal is to register anatomical and functional brain images of the same patient to facilitate the process of functional localization. This non-rigid image registration of different modalities makes possible to obtain a geometric correspondence which allows for localizing the functional processes that occur in the brain. In order to evaluate the performance of the proposed method, visual and numeric results of registration are shown. The quality of the registration results is measured by considering the peak signal to noise ratio (PSNR), the mutual information (MI) and the correlation ratio (CR).

Content analysis of pollen grains in the atmosphere is an important task for preventing allergy symptoms, studying crop production or detecting environmental changes. In the last decades, a lot of palynological labs have been created to collect, prepare and analyse airborne samples. Nowadays, this task is done manually with optical microscopes, requires trained experts and is time-consuming. The development of new computer vision systems and the low price of storage systems have improved the solutions towards an automated palynology. Some recognition problems have been solved with better quality images and other with 3D images, but localisation in real airborne samples, with debris, clumped and grouped pollen grains needs to be improved in order to achieve an automatic system useful for biological labs. In this manuscript, we analyse the advances achieved in the last years and explain a new low-cost methodology, that imitates the human expert labour using computational algorithms based on image characteristics and domain knowledge to detect pollen grains. The current results are promising (81.92% of recall and 18.5% of precision) but not enough to develop an automated palynology system.

This paper addresses a fully automatic segmentation method for ultrasound images of the common carotid artery. The goal of this procedure is the detection of the arterial walls to assist in the evaluation of the arterial diameter. In other words, the main objective is the segmentation of the region corresponding to the lumen of the vessel, where the blood flows. The evaluation of the Lumen Diameter (LD) provides useful information for the diagnosis of arterial diseases. The monitoring of LD and Intima-Media Thickness (IMT) is crucial in the early detection of atherosclerosis and in the assessment of the cardiovascular risk. The proposed methodology is completely based on Machine Learning and it applies Auto-Encoders and Deep Learning to obtain abstract and efficient data representations. Thus, the segmentation task is posed as a pattern recognition problem. The different architectures designed have shown a good classification performance. In addition, the results obtained for some ultrasound images of the common carotid artery can be visually validated in this work. The final automatic segmentation is quite accurate, and it is possible to conclude that it will lead to a precise and reliable measurement of the lumen diameter.

The effects of epilepsy in a patient can be significantly reduced with medical treatment. However, in some patients or after some time the anti-epileptic drugs do not take effect, being candidates to surgery. Preliminary studies of the patient are usually limited to EEG and MRI, and the epileptic focus is located using brain imaging algorithms that do not provide enough certainty to the specialist. In this work four of the most widely used free distribution neuroimaging software are tested with real epileptic data (EEGLab, SPM, LORETA, and Cartool), with the objective of illustrating their capabilities for locating the epileptic focus. As a result, a novel methodology for robust estimation that includes the advantages of the four software is proposed.

A novel Weighted Unscented Kalman Filtering method is introduced for neural activity estimation from electroencephalographic signals. The introduction of a weighting stage improves the solution by extracting relevant information directly from the measured data. Besides, a discrete nonlinear state space model representing the brain neural activity is used as a physiological constraint in order to improve the estimation. Moreover, time-varying parameters are considered which allow describing adequately healthy and pathological activity even for localized epilepsy events. Performance of the new method is evaluated in terms of introduced error measurements by application to simulated EEG data over several noise conditions. As a result, a considerable improvement over linear estimation approaches is found.

Estimation of neural activity using Electroencephalography (EEG) signals allows identifying with high temporal resolution those brain structures related to pathological states. This work aims to improve spatial resolution of estimated neural activity employing time-varying dynamic constraints within the iterative inverse problem framework. Particularly, we introduce the use of Dynamic Neural Fields (DNF) to represent neural activity directly related to epileptic foci localization adequately. So, we develop a DNF-based time variant estimation model in the form of an Iterative Regularization Algorithm (IRA) that carries out neural activity estimation at every time EEG sample. The IRA model performance that is evaluated on simulated and real cases is compared with the baseline static and dynamic methods under several noise conditions. To this end, we use different error measures showing that the IRA estimation model can be more accurate and robust than the other compared methods.

Many medical applications commonly make use of brain magnetic resonance images (MRI) as an information source since they provide a non-invasive view of the head morphology and functionality. Such information is given by the properties of head structures, which are extracted using segmentation techniques. Among them, multi-atlas-based methodologies are the most popular, allowing to consider prior spatial information about the distribution of brain structures. These approaches rely on a non-linear mapping of the information of the most relevant atlases to a query image. Nevertheless, methodology effectiveness is highly dependent on the mapping function and the atlas relevance criterion, being both of them based on the selection of an MRI similarity metric. Here, a new spatially weighting measure is proposed to enhance the multi-atlas-based segmentation results. The proposal is tested in an MRI segmentation database for state-of-the-art image metrics as means squares, histogram correlation coefficient, normalized mutual information, and neighborhood cross-correlation and compared against other spatial combination approaches. Achieved results show that our proposal outperforms baseline methods, providing a more suitable atlas selection.

We investigate the use of the iLU preconditioning within the framework of the Anisotropic-Finite-Difference based Solution for the EEG Forward Problem. Provided the minimal error of representation, comparison of the convergence rate and computational cost is carried out for several competitive numerical solver combinations. From the testing on real data, we obtain that combination of the biconjugate gradient solver and incomplete LU factorization results in a numerical solution that outperforms the other considered approaches in terms of accuracy and computational cost. We validate this numerical solution combination against analytical spherical mode. Also, testing on realistic head models (with high anisotropic areas and heterogeneous tissue conductivities) shows high accuracy and low computational cost.

The waveform of physiological signals carries useful information about the brain states. Automated computational algorithms are used in clinical medicine for extracting this information that cannot be read directly by visual inspection. Nonetheless, difficulties arise in the extraction because the intrinsic rhythms of the waveforms vary with the changes in the state of the brain. That is the case for electroencephalogram (EEG) signals from Epileptic seizure events. Here, we address the extraction of information from EEG signals by using a novel methodology that quantitatively measures the intrinsic rhythms of EEG waveforms related to healthy or Epileptic seizure events. In this method, the customized wavelet is used to estimate the EEG rhythms and then the relevance analysis with Fuzzy entropy and Stochastic measure are used to discriminate between seizure free and seizure states. The classification stage is based on classification performance using a support vector machine classifier. The pertinence of the proposed methodology during the Epileptic seizure identification is discussed, and future directions are presented.

Based on the assumption that brain activity appears in localized brain regions that can vary along time, yielding spatial and temporal non-stationary activity, we propose a constrained M/EEG inverse solution, based on the Fused Lasso penalty, that reconstructs brain activity as dynamic small and locally smooth spatial patches. Thus, our main contribution is to provide neural activity reconstruction tracking non-stationary dynamics. We validate the proposed approach in two different ways: i) using simulated MEG data when we have previous knowledge about spatial and temporal signal dynamics, and ii) using real MEG data, particularly we use a faces perception paradigm aimed to examine the M170 response. In the former case of validation, our approach outperforms conventional M/EEG-based imaging algorithms. Besides, there is a high correspondence between brain activities presented on the evaluated real MEG data and the time-varying solution obtained by our approach.

The analysis of coactive regions during a Motor Imagery (MI) task becomes an important issue for revealing the primary neural activity provided by movement intentions. This information should be useful in the design of Brain Computer Interface systems. In this work, a connectivity analysis strategy for the MI paradigm using short-time features and kernel similarities is proposed. Since the imagination and execution of tracking movements are associated with neural rhythm power changes in the

μ

and

β

bands, we estimate three representative short-time feature extraction methods (Power spectral density, Hjort, and wavelet parameters). Moreover, a kernel-based pairwise similarity is computed among channels to highlight brain coactive areas during a MI task. In addition, the influence of an EEG preprocessing stage before computing the short-time features and the similarity among channels is studied. The attained results demonstrate that our approach can capture the main brain activity relationships in accordance with the MI paradigm clinic findings.

This paper falls under the idea of introducing biomimetic miniature air vehicles in ambient assisted living and home health applications. The concepts of active disturbance rejection control and flatness based control are used in this paper for the trajectory tracking tasks in the flapping-wing miniature air vehicle (FWMAV) time-averaged model. The generalized proportional integral (GPI) observers are used to obtain accurate estimations of the flat output associated phase variables and of the time-varying disturbance signals. This information is used in the proposed feedback controller in (a) approximate, yet close, cancelations, as lumped unstructured time-varying terms, of the influence of the highly coupled nonlinearities and (b) the devising of proper linear output feedback control laws based on the approximate estimates of the string of phase variables associated with the flat outputs simultaneously provided by the disturbance observers. Numerical simulations are provided to illustrate the effectiveness of the proposed approach.

Even in its early stages, the cognitive deficits in persons with dementia (PwD) can produce significant functional impairment. Dementia is characterized by changes in personality and behavioral functioning that can be very challenging for caregivers and patients. This paper presents results on the use and adoption of a cognition assistive system to support occupational therapy to address psychological and behavioral symptoms of dementia. During 6 months we conducted an in situ system evaluation with a caregiver-PwD dyad to evaluate the adoption and effectiveness of the system to ameliorate challenging behaviors. Evaluation results indicate that intervention personalization and touch-based systems interfaces encouraged the adoption and the positive effect in reducing challenging behaviors in PwD and decreases caregiver burden.

The syndrome of hemispatial neglect is usually associated to a lesion of the brain and is characterized by a reduced or lack of awareness of one side of space, even though there may be no sensory loss. Although it is extremely common, it has proven to be a challenging condition both to understand and to treat. This paper focuses on reviewing this syndrome and proposing new therapies based on multisensory feedback in a virtual environment. These therapies have been designed to improve the awareness of the neglected side by using visual, auditory and haptic feedback.

This paper introduces a study on the relationship between emotion regulation and color preference. In the described pilot study, participants are asked to label uniform color images by using opposite meaningful words belonging to four semantic scales, namely “Tension” (ranging from

Relax

to

Stress

), “Temperature” (

Coldness

to

Warmness

), “Amusement” (

Boredom

to

Fun

) and “Attractiveness” (

Pleasantness

to

Unpleasantness

). Simultaneously, the participants have to indicate if they feel certain emotions while observing each colored image, as well as to rate the intensity of the feeling. The labeled emotions are “Joy”, “Happiness”, and “Sadness”. The results demonstrate that people generally perceive color emotions for one-colored images in similar ways, though showing some variations for males and females. Several conclusions about the relations between color and emotions are presented.

This article is based on the assumption of the power of music to change the listener’s mood. The proposal studies the participants’ changes in emotional states through listening different auditions. This way it is possible to answer to the question if music is able to induce positive and negative emotions in the listener. The present research focuses on the musical parameter of note value through its four basic components of the parameter note value, namely, beat, rhythm, harmonic rhythm and rhythmic accompaniment to detect the individual preferences of the listeners. The initial results prove that the influence of beat in music for eliciting emotions is dependent of the personality of each participant in terms of neuroticism and extraversion.

In this paper, we enhance systems interacting in healthcare domains by means of incorporating emotionally sensitive spoken conversational interfaces. The emotion recognizer is integrated in these systems as an intermediate phase between natural language understanding and dialog management in the architecture of a spoken dialog system. The prediction of the user’s emotional state, carried out for each user turn in the dialog, makes it possible to adapt the system dynamically selecting the next system response taking into account this valuable information. We have applied our proposal to develop an emotionally sensitive conversational system adapted to patients suffering from chronic pulmonary diseases, and provide a discussion of the positive influence of our proposal in the perceived quality.

This proposal is framed within the group’s general working line of applying artificial intelligence techniques to advance in early mild cognitive impairment diagnosis. If impairment in semantic production was studied in previous works, now we rely on the reduced ability to reproduce or copy simple figures, part of standardized neuropsychological tests designed to assess mild cognitive impairment. Although the long-term goal of this project is to work with all figures from these tests, in this paper we will focus on the automatic analysis of the alternating graphs figure. We develop a quantitative descrition of different features that appear to be very abstract in the test norms and define new features that are not considered so far. Results with just one figure are quite promising (77.7% precision and 77.1 recall).

Loitering is a common behaviour of the elderly people. We goal is develop an artificial intelligence system that automatically detects loitering behaviour in video surveillance environments. The first step to identify this behaviour was used a Generalized Sequential Patterns that detects sequential micro-patterns in the input loitering video sequences. The test phase determines the appropriate percentage of inclusion of this set of micro-patterns in a new input sequence, namely those that are considered to form part of the profile, and then be identified as loitering. The system is dynamic; it obtains micro-patterns on a repetitive basis. During the execution time, the system takes into account the human operator and updates the performance values of loitering in shopping mall. The profile obtained is consistent with what has been documented by experts in this field and is sufficient to focus the attention of the human operator on the surveillance monitor.

As the population increases in the world, the ratio of health carers is rapidly decreasing. Therefore, there is an urgent need to create new technologies to monitor the physical and mental health of people during their daily life. In particular, negative mental states like depression and anxiety are big problems in modern societies, usually due to stressful situations during everyday activities including work. This paper presents a machine learning approach for stress detection on people using wearable physiological sensors with the final aim of improving their quality of life. The presented technique can monitor the state of the subject continuously and classify it into ”stressful” or ”non-stressful” situations. Our classification results show that this method is a good starting point towards real-time stress detection.

Millions of elderly people around the world use the walker for their mobility; nevertheless, these devices may lead to an accident. One of the cause of these accidents is misjudge the terrain. The main objective of this work is the implementation of a ground change detector in real time on a small and light embedded system that can be clipped on a rollator. As a long-term goal, this device will allow users to anticipate entering dangerous situations. We implemented an algorithm to detect ground changes based on color histograms and texture descriptor given as inputs to multi-layer perceptrons. Experiments were performed both off-line and with an embedded system. The obtained results indicated that it is possible to have an accurate detector which is able to distinguish ground changes in real-time.