Quantifying and Processing Biomedical and Behavioral Signals

Currently, researchers coming from psychological, computational, and engineering research fields have developed a human-centered behavioral informatics characterized by techniques analyzing and coding human behaviors, conventional and unconventional social conducts, signals coming from audio and video recordings, auditory and visual pathways, neural waves, neurological and cognitive disorders, psychological and personal traits, emotional states, mood disorders. This interweaving of expertise had produced extensive research progresses and unexpected converging interests allowing the groundwork for a book dedicated to pose the current progresses in dynamics of signal exchanges and reporting the latest advances on the synthesis and automatic recognition of human interactional behaviors. Key features considered are the fusion and implementation of automatic processes and algorithms for interpreting, tracking, and synthesizing dynamic signals such as facial expressions, gaits, EEGs, brain and speech waves. The acquisition, analysis, and modeling of such signals is crucial for computational studies devoted to a human-centered behavioral informatics.

Neurocognitive self-enhancement can be defined as a voluntary attempt to improve one’s own cognitive skills and performance by means of neuroscience techniques able to influence the activity of neural structures and neural networks subserving such skills and performance. In the last years, the strive to improve personal potential and efficiency of cognitive functioning lead to the revival of mental training activities. Recently, it has been suggested that such practices may benefit from the support of mobile computing applications and wearable body-sensing devices. Besides discussing such topics, we report preliminary results of a project aimed at investigating the potential for cognitive-affective enhancement of a technology-mediated mental training intervention supported by a novel brain-sensing wearable device. Modulation of motivational and affective measures, neuropsychological and cognitive performances, and both electrophysiological and autonomic reactivity have been tested by dividing participants into an experimental and an active control group and by comparing the outcome of their psychometric, neuropsychological, and instrumental assessment before, halfway through, and after the end of the intervention period. The technology-mediated intervention seemed to help optimizing attention regulation, control and focusing skills, as marked by a reduction of response times at challenging computerized cognitive tasks and by the enhancement of event-related electrophysiological deflections marking early attention orientation and cognitive control. Available evidences, together with the first set of findings here reported, are starting to consistently show the potential of available methods and technologies for enhancing human cognitive abilities and improving efficiency of cognitive processes.

This paper investigates the ability of adolescents (aged 13–15 years) and young adults (aged 20–26 years) to decode affective bursts culturally situated in a different context (Francophone vs. South Italian). The effects of context show that Italian subjects perform poorly with respect to the Francophone ones revealing a significant native speaker advantage in decoding the selected affective bursts. In addition, adolescents perform better than young adults, particularly in the decoding and intensity ratings of affective bursts of happiness, pain, and pleasure suggesting an effect of age related to language expertise.

This study investigates the valence of adults’ implicit associations to typical and atypical infant cues, and the consistency of responses across the different stimuli. 48 non-parent adults (25 females, 23 males) were presented three kinds of infant cues, typical cry (TD-cry), atypical cry (ASD-cry) and infant faces, and their implicit associations were measured by means of the Single Category Implicit Association Test (SC-IAT). Results showed that, independently of gender, the implicit associations to typical and atypical infant cries had the same negative valence, whereas infant faces were implicitly associated to the positive dimension. Moreover, data showed that implicit responses to the different infant cues were not associated. These results suggest that more controlled processes influence the perceptions of atypical infant cry, and confirm the need to investigate individual reactions to infant cues by adopting a multilevel approach.

Starting from the assumption that caregiving behaviours are regulated at different levels, the aim of the present paper was to investigate adults’ reaction to salient infant cues by means of a multilevel approach. To this aim, psychophysiological responses (Heart Rate Variability), implicit associations (SC-IAT-A), and explicit attitudes (semantic differential) toward salient infant cues were measured on a sample of 25 non-parents adults (14 females, 11 males). Moreover, the trait anxiety and the individual noise sensitivity were considered as controlling factors. Results showed that adults’ responses were moderated by the specific measure considered, and that responses at the different levels were only partially consistent. Theoretical and practical implications were discussed.

Olfactory perception is affected by cross-modal interactions between different senses. However, although the effect of cross-modal interactions for smell have been well investigated, little attention has been paid to the facilitation expressed by haptic interactions with a manipulation of the odorous object’s shape. The aim of this research is to investigate whether there is a cortical modulation in a visual recognition task if the stimulus is processed through an odorous cross-modal pathway or by haptic manipulation, and how these interactions may have an influence on early visual-recognition patterns. Ten healthy non-smoking subjects (25 years ± 5 years) were trained to have a haptic manipulation of 3-D models and olfactory stimulation. Subsequently, a visual recognition task was performed during an electroencephalography recording to investigate the P3 Event Related Potentials components. The subjects had to respond on the keyboard according to their subjective predominant recognition (olfactory or haptic). The effects of haptic and olfactory condition were assessed via linear mixed-effects models (LMMs) of the lme4 package. This model allows for the variance related to random factors to be controlled without any data aggregation. The main results highlighted that P3 increased in the olfactory cross-modal condition, with a significant two-way interaction between odor and left-sided lateralization. Furthermore, our results could be interpreted according to ventral and dorsal pathways as favorite ways to olfactory crossmodal perception.

In order to provide support to the implementation of on-line and remote systems for the early detection of interactional disorders, this paper reports on the exploitation of handwriting and drawing features for detecting negative moods. The features are collected from depressed, stressed, and anxious subjects, assessed with DASS-42, and matched by age and gender with handwriting and drawing features of typically ones. Mixed ANOVA analyses, based on a binary categorization of the groups, reveal significant differences among features collected from subjects with negative moods with respect to the control group depending on the involved exercises and features categories (in time or frequency of the considered events). In addition, the paper reports the description of a large database of handwriting and drawing features collected from 240 subjects.

Music has an extraordinary ability to evoke emotions. Nowadays, the music fruition mechanism is evolving, focusing on the music content. In this work, a novel approach for agglomerating songs on the basis of their emotional contents, is introduced. The main emotional features are extracted after a pre-processing phase where both Sparse Modeling and Independent Component Analysis based methodologies are applied. The approach makes it possible to summarize the main sub-tracks of an acoustic music song (e.g., information compression and filtering) and to extract the main features from these parts (e.g., music instrumental features). Experiments are presented to validate the proposed approach on collections of real songs.

Sport, in different ways, change considerably people’s life. The purpose of this experiment was to reveal possible association between the stimulation of sympathetic system induced by exercise and the one induced by the rise of systemic concentration of Orexin A and bring the truth about orexins and sport network. Blood samples were collected from subjects (men, n = 10; age: 23.2 ± 2.11 years) 15, 0 min before the start of exercise, and 30, 45, 60 min after a cycle ergometer exercise at 75 W for 15 min. Also, heart rate (HR), galvanic skin response (GSR), and rectal temperature were monitored. The exercise produce a significant rise (p < 0.01) in plasmatic orexin A with a peak at 30 min after the exercise bout, in association with a rise of the other three monitored variables: HR (p < 0.01), GSR (p < 0.05), and rectal temperature (p < 0.01). Our results indicate that plasmatic orexin A is involved in the reaction to physical activity and in the beneficial effects of sport.

In this paper, we analyze in-air movements in online handwriting databases when the distance from the tip of the pen to the paper surface is higher than 1 cm. In this case, the computer can only know the time spent in air because the distance is too high to track the x and y coordinates of the movement. While this kind of movement is usually discarded, some investigation must be done in order to decide if computational algorithms can take advantage of this information in some scenarios. In this paper, we establish a criterion to differentiate useful in-air long distance strokes from user pauses.

Inertial sensors for analysing some biomechanics conditions have long been studied in the medical and sports fields. In order to improve any qualitative assessment related to detecting the degree of injury and the range of motion in the cervical area, the system needs to be robust enough. It is important to detect purposefully altered data, reduce subjective variables effect (such as pain or discomfort) and accurately determine the impairment or dysfunction levels. The first aim of this work was to produce a multi-sensor database for the cervical area by gathering data from an inertial system, from an EEG head-set and from a thermographic camera. This complete set of information can provide further insight when researchers try to develop an objective diagnostic algorithm or improve intelligent diagnostic systems.

Historically, the relation between consciousness and the brainstem has been demonstrated, on the one hand, by injuries to the upper brainstem that lead to minimum states of consciousness, comas and persistent vegetative states and, on the other hand, by electrophyisiological recordings that link the ascending reticular activating system (ARAS) with vigilance and attention, functions which are necessary for interpersonal relationships. With the advances made in the clarification of the connections between the brainstem and other regions of the brain there has been no corresponding conceptual revision of the functional context within which the ARAS performs its role of activation and way in which it activates the cerebral cortex, unlike other structures of the brain. In this paper we shall discuss the way in which the brainstem—(a) fundamental terminal of multiple ascending neural pathways—influences and modulates cortical activities; (b) the context of which the ARAS is a fundamental part—the centroencephalic archipelago of functional integration—for the transmission of contents to specific regions that generate the sensation of subjectivity.

The systematic study of biological basis of behavior and of the process involved in economical choices has outlined a new paradigm of research: neuroeconomics. Now the intersection between neuroscience, psychology and economics, neuroeconomics presents itself as an alternative to the neoclassical vision on economics, according to which the homo oeconomicus acts within the bonds of a formalizing rationality tending to the maximization of the anticipated utility. Brain imagining methods have shown that the decision-making processes activate the frontal lobe and the limbic system above all, a big circonvolution running through the callous body on the medial surface of the hemispheres, extending itself down, responsible for the regulation of emotional phenomena. Reinforcing such a tendency, we find the injury paradigm. It was observed that frontal lobe injuries harm the capacity of making advantageous decisions either in one’s own behalf or in others, as well as decisions according to the social conventions. In this paper, we will try to show that if, by the one hand, the neuro visual methods have given us a great amount of data, on the other hand, using them uncritically, with the recurrent confusion between “correlation” and “causal relation”—contemporary microevents indicate only simple correlations, and no cause-effect relation—risks to stress the relevant explanatory gap regarding the abstract ideal of understanding the nature of the brain.

This paper addressed the issue of epileptic absence seizures developing a complex brain network model based on the estimation of the coherence between electroencephalographic (EEG) signals. The EEG signals indeed reflect the abnormalities in the cortical electrical activity caused by epilepsy. A dataset of 10 absence patients was analyzed, including 63 seizures. The model was analyzed over the time to assess if changes in the network parameters matched the brain state (ictal: seizure), (non-ictal: seizure free). During the ictal states, the characteristic path length (\(\lambda \)) decreased and the global efficiency (GE), the average clustering coefficient (CC) and the small worldness (SW) increased, as expected, because of the abnormal synchronization associated with absence seizure onset. The connection matrices preceding the ictal states (8 s before) were thresholded and the corresponding connectivity scalp maps, showing the active links between EEG channels, were displayed. Such connectivity maps showed the interaction between channels and provided information about the abnormal recruitment mechanism associated with seizure development: the involvement of the cortical areas appears progressive and that every subject exhibited peculiar recurrent patterns of area activation.

Alzheimer’s disease (AD) is the most common neurodegenerative disorder characterized by cognitive and intellectual deficits and behavior disturbance. The electroencephalogram (EEG) has been used as a tool for diagnosing AD for several decades. In the pre-clinical stage of AD, no reliable and valid symptoms are detected to allow a very early diagnosis. There are four different stages associated with AD. The first stage is known as Mild Cognitive Impairment (MCI), and corresponds to a variety of symptoms which do not significantly alter daily life. In the mild stage, an impairment of learning and memory is usually notable. The next stages (Mild and Moderate AD) are characterized by increasing cognitive deficits and decreasing independence, culminating in the patient’s complete dependence on caregivers and a complete deterioration of personality (Severe AD). In this paper, we propose the study of the evolution of Alzheimer’s disease using eLORETA’s three-dimensional distribution of the current density and Small-world network. Our goal is to see the changes of MCI patients’ EEG (called EEG T₀) after three months (EEG T₁). The results show that small-world is a valid technique to see the temporal evolution of the disease.

The paper presents a dynamic gesture recognizer, that assumes that the gesture can be described by Kendon Gesture model. The gesture recognizer has four modules. The first module performs the feature extaction, using the skeleton representation of the body person provided by NITE library of Kinect. The second module, formed by Learning Vector Quantization, has the task of individuating the initial and the final handposes of the gesture, i.e., when the gesture starts and terminates. The third unit performs the dimensionality reduction. The last module, formed by a discrete Hidden Markov, perfoms the gesture classification. The proposed recognizer compares favourably, in terms of accuracy, most of existing dynamic gesture recognizers.

Image recognition applications has been capturing interest of researchers for many years, as they found countless real-life applications. A significant role in the development of such systems has recently been played by evolutionary algorithms. Among those, HyperNEAT shows interesting results when dealing with potentially high-dimensional input space: the capability to encode and exploit spatial relationships of the problem domain makes the algorithm effective in image processing tasks. In this work, we aim at investigating the effectiveness of HyperNEAT on a particular image processing task: the automatic segmentation of blood vessels in retinal fundus digital images. Indeed, the proposed approach consists of one of the first applications of HyperNEAT to image processing tasks to date. We experimentally tested the method over the DRIVE and STARE datasets, and the proposed method showed promising results on the study case; interestingly, our approach highlights HyperNEAT capabilities of evolving towards small architectures, yet suitable for non-trivial biomedical image segmentation tasks.

In the past few years, several works describing systems for the prompt detection of falls have been presented in literature. Many of these systems address the problem of fall detection by using some handcrafted features extracted from the input signals. In the meantime interest in the use of feature learning and deep architectures has been increasing, thus reducing the required engineering effort and the need for prior knowledge. A fall detection method based on a Deep Convolutional Neural Network Autoencoder is presented in this work. This method is trained as a novelty detector through the end-to-end strategy. The classifier distinguishes normal sound events generated by common indoor human activity (i.e. footsteps and speech) and music background from novelty sound events produced by human falls. The performance of the algorithm has been assessed on a corpus of fall events created by the authors. Moreover a comparison was made with two different state-of-art algorithms both based on a One Class Support Vector Machine. The results showed an improvement on performance of about 11% on average.

This work reports on a study of web usage logs to verify whether it is possible to achieve good recognition rates in the task of distinguishing between human users and automated bots using computational intelligence techniques. Two problem statements are given, offline (for completed sessions) and on-line (for sequences of individual HTTP requests). The former is solved with several standard computational intelligence tools. For the second, a learning version of Wald’s sequential probability ratio test is used.

A main question in urban environments is the continuous growth of private mobility with its negative effects such as traffic congestion and pollution. To mitigate them, it is important to promote different forms of mobility among the citizens. Car-sharing systems give users the same flexibility and comfort of private cars but at smaller costs. For this reason, car-sharing has continuously increased its market share although rather slowly. To boost such growth, car-sharing systems needs to increase vehicle fleet, improve company profits and, at the same time, make it more affordable for consumers. In this paper the promotion of car-sharing by reputation is proposed. Neural networks have been used to identify drivers’ habits in using car-sharing vehicles. To verify the effectiveness of the proposed approach, some experiments based on real and simulated data were carried out with promising results.

Cancer is a large family of genetic diseases that involve abnormal cell growth. Genetic mutations can vary from one patient to another. Therefore, personalized precision is required to increase the reliability of prognostic predictions and the benefit of therapeutic decisions. The most important issues concerning gene analysis are strong noise, high dimensionality and minor differences between observations. Therefore, it has been chosen an unsupervised approach in order to bypass the high dimensionality issue using parallel coordinates and a scoring algorithm of features based on their clustering ability. Traditional methods of dimensionality reduction and projection are here used on subset features with high discriminant power in order to better analyze the data manifold and select the more meaningful genes. Previous studies show that mutations of genes NRAS, KRAS and BRAF lead to a dramatic decrease in therapeutic effectiveness. The following analysis tries to explore in an unconventional way gene expressions over tissues which are wild type regarding to these genes.

Nowadays, uterine fibroids can be treated using Magnetic Resonance guided Focused Ultrasound Surgery (MRgFUS), which is a non-invasive therapy exploiting thermal ablation. In order to measure the Non-Perfused Volume (NPV) for treatment response assessment, the ablated fibroid areas (i.e., Region of Treatment, ROT) are manually contoured by a radiologist. The current operator-dependent methodology could affect the subsequent follow-up phases, due to the lack of result repeatability. In addition, this fully manual procedure is time-consuming, considerably increasing execution times. These critical issues can be addressed only by means of accurate and efficient automated Pattern Recognition approaches. In this contribution, we evaluate two computer-assisted segmentation methods, which we have already developed and validated, for uterine fibroid segmentation in MRgFUS treatments. A quantitative comparison on segmentation accuracy, in terms of area-based and distance-based metrics, was performed. The clinical feasibility of these approaches was assessed from physicians’ perspective, by proposing an integrated solution.

Cancer is a large family of genetic diseases that involve abnormal cell growth. Genetic mutations can vary from one patient to another. Therefore, personalized precision is required to increase the reliability of prognostic predictions and the benefit of therapeutic decisions. The most important issues concerning gene analysis are strong noise, high dimensionality and minor differences between observations. Therefore, parallel coordinates have been also used in order to better analyze the data manifold and select the more meaningful genes. Later, it has been chosen to implement a supervised feature selection algorithm in order to work on a subset of features only avoiding the high dimensional problem. Other traditional methods of dimensionality reduction and projection are here used on subset features in order to better analyze the data manifold and select the more meaningful gene. Previous studies show that mutations of genes NRAS, KRAS and BRAF lead to a dramatic decrease in therapeutic effectiveness. The following analysis tries to explore in an unconventional way gene expressions over tissues which are wild type regarding to these genes.

This paper proposes a methodology to assess the discriminative capabilities of geometrical descriptors referring to the public Bosphorus 3D facial database as testing dataset. The investigated descriptors include histogram versions of Shape Index and Curvedness, Euclidean and geodesic distances between facial soft-tissue landmarks. The discriminability of these features is evaluated through the analysis of single block of features and their meanings with different techniques. Multilayer perceptron neural network methodology is adopted to evaluate the relevance of the features, examined in different test combinations. Principle component analysis (PCA) is applied for dimensionality reduction.

This paper presents a comparison between two different Computer Aided Diagnosis systems for classification of five types of leucocytes located in the tail of a Peripheral Blood Smears: Lymphocytes, Monocytes, Neutrophils, Basophils and Eosinophils. In particular, we have evaluated and compared the performance of a previous feature-based Back Propagation Neural Network classifier with the performance of two novel classifiers both based on Deep Learning using Convolutional Neural Networks introduced in this study. All the classifiers are built considering the same dataset of images acquired in a previous study. The experimental results, reported in terms of accuracy, sensitivity, specificity and precision, show that the different strategies could be compared and discussed from both clinical and technical point of view.