The gustatory areas of the brain include the primary (GI) and the secondary (GII) gustatory cerebral cortex. A spatial mechanism has been recently hypothesized to explain the taste quality recognition. This research investigates the spatial distribution of blood oxygen level dependent (BOLD) activations evoked in the human area GI by different tastants, aimed at building a chemotopic map. The chemotopic organization of the human GI was studied in seven healthy subjects by applying three taste stimuli (salty, sweet, neutral) to either side of the tongue, using a 5-min functional magnetic resonance imaging (fMRI) block-designed protocol, alternating periods of rest and stimulation. Data were analyzed by the Functional Magnetic Resonance Imaging of the Brain (FMRIB) Software Library (FSL). Unilateral tongue stimulation consistently evoked bilateral activation in area GI. Ipsilateral foci were generally larger and signal increases greater. The foci evoked by each tastant exhibited slightly but not significantly different mean coordinates, broad overlap and high interindividual variability; the salty stimuli generally evoked more anterior foci and sweet stimuli more posterior activation. Results confirm that the gustatory pathways from tongue to cortex are bilaterally distributed, with an ipsilateral predominance. Although distinct GI zones were activated by the different taste stimuli, a clear topographical organization could not be recognized. Possibly, fMRI technique is unable to resolve fine topographical arrangements or GI discriminative role for different tastants is subserved by another mechanism. Bilateral activation of the primary somatosensory area in the parietal cortex (contralateral predominance) and of middle insula (ipsilateral predominance) were also observed.

Modern rehabilitation and assistive devices require the use of smart interfaces able to capture the subject's intent of motion and translate such intent to specific control strategies. The use of surface electromyography (sEMG) signals, in combination with data-driven models, constitutes a viable framework to solve the aforementioned problem. Although literature highlighted the tendency toward a multiple sensors approach, a minimal set-up may reduce costs and complexity of myoelectric interfaces. In this study, gastrocnemius lateralis (GAL) and tibialis anterior (TA) sEMG signals were used in order to investigate their single and combined role in the flexion-extension angles estimation of ankle and knee during gait. Least-square support vector machine (LS-SVM) with linear, polynomial, and radial basis function (RBF) kernel was employed to estimate the most suitable function that maps the myoelectric information from single muscle and from the combination of both, in lower limbs joint kinematics. LS-SVM with RBF outperformed the other kernels in the ankle and knee kinematics estimation for all the 6 subjects examined. Moreover, when using RBF with the only GAL data, the median root mean square error (RMSE) values were above 5 degrees for ankle and 8 degrees for knee angles whereas the combined information from GAL and TA showed slightly better results. Outcomes support a minimal electrodes set-up for the development of lower limb myoelectric interfaces for kinematic estimation.

Physical activity is recommended in patients with type 1 diabetes (T1D), but therapy management still lacks efficient tools to avoid exercise-induced hypoglycemia. Machine learning represents a powerful solution in the field of decision support for diabetes management and its application to continuous glucose monitoring (CGM) data appears promising in pre-exercise prediction of upcoming adverse events. Aim of this study was to investigate the possibility to distinguish if a specific configuration of CGM metrics evaluated before starting of exercise is more prone to induce hypoglycemia after the start of the exercise session until the following day. A total of 47 CGM recordings from T1D children have been used to extract CGM metrics from pre-exercise CGM data. Acquisitions were labelled as HYPO or as NO-HYPO, respectively if belonging to subjects who experienced or did not experience hypoglycemia during the time following the exercise. Anthropometric characteristics and extracted features have been given as input to a decision tree classification algorithm to select those with the most predictive power. The selected features were then further evaluated with respect to the classification problem by using them as input to other three classification models: random forest, adaboost and gradient boosting. Performance results in terms of area under receiver operating characteristic (AUC) were as follows: 85.5%, 82.1%, 78.1% and 74.3% for decision tree, gradient boosting, random forest and adaboost, respectively. M-value, maximum glucose, time above 180 mg/dL and time above 250 mg/dL could have a role in predicting upcoming hypoglycemia prior the starting of exercise.

During physical exercise, the assessment of cardiac autonomic regulation through the analysis of heart-rate variability is difficult due to non-stationarity of RR intervals. The short duration of stationary epoch of RR-interval series makes not possible the computation of classical time-and frequency-domain parameters. Symbolic analysis can deal with short epoch of RR interval; thus, this study aims to apply symbolic analysis to analyze heart-rate variability of short-distance runners during training and competition. Data consists of RR-intervals extracted from 30s-long electrocardiograms acquired by KardiaMobile of Alivecor in 8 short-distance runners (1/7 M/F, 17[16;20] years) during training and competition. Symbolic analysis classifies a reduced number (in this study, three) of consecutive RR intervals in four patterns according to the sign and number of variations: no variation (0V); one variation (1V); two like variations (2LV); two unlike variations (2UV). An increase in low amount of variations (0V or 1V patterns) is usually linked with increased sympathetic control and vagal withdrawal, while an increase in high amount of variations (2LV and 2UV patterns) is usually linked with sympathetic withdrawal and increased vagal control. In our results, pattern 0V /2LV increases/decreases from rest to post exercise and decreases/increases during recovery, in both training and competition. Thus, our results confirm the opposite trends between low and high variations of symbolic patterns. In conclusion, symbolic analysis seems to be an efficient tool to characterize heart-rate variability during physical exercise at different level of psychophysical stress.

The study of motor disorders due to neurodegenerative diseases (NDD) is assuming a central role in healthcare systems, this is certainly due to the needs of early recognition systems that can allow a better management of the patients daily-life. Many studies in the literature faced the problem of finding digital biomarkers from data collected through gait experiments to discriminate between control (CN) and NDD groups without systematically face the problem of which gait time-series were more appropriate to extract opportune descriptors for characterizing the NDD considered. In this work, such problem was modeled through a machine learning approach. Thus, 6 time-dependent spectral features (PSDTD) were extracted from 4 gait time-series, i.e., stride (SR), stance (SA), swing (SW) and double support (DS) duration intervals. A publicly available data set containing data of CN, Parkinson's (PD), Huntington's (HD) and amyotrophic lateral sclerosis (ALS) diseases was employed to the purpose. Low error rates using leave one out validation scheme were obtained using PSDTD features computed over DS and SA for CN-PD and CN-HD classification, i.e., error rate < 0.1 for DS and < 0.15 for SA. Regarding CN-ALS classification, best results were obtained using SA features, i.e. error rate <0.07. This supports the research line that dynamic equilibrium phases of the gait can hide important biomarkers for the characterization of different NDD.

Alzheimer's Disease (AD) is the most common neurodegenerative disease. Its first stage, namely prodromal or Mild Cognitive Impairment (MCI), is characterized by slightly structural changes in the subcortical structures of the temporal lobe. Brain Magnetic Resonance (MR) is the most utilized neu-roimaging modality for the diagnosis of AD. Although an early therapeutic intervention during the initial stages of AD appears to have a positive impact on the progression of symptoms, its accurate diagnosis is still very difficult. Deep Learning (DL)-based decision-support systems hold great potential in generalizing even under subtle anatomical changes of the brain, like the ones caused by AD at its onset. To our knowledge, we were the first to develop a Convolutional Long Short-Term Memory (ConvLSTM)-based decision-support system and an improved version of it for the automatic diagnosis of AD from 3D brain MR. The research presented in this paper aims to extend their applicability to MCI for effectiveness verification through the development of CLAUDIA, a new on-cloud decision-support system for the automatic diagnosis of Alzheimer's prodromal stage and disease from 3D brain MR. To this aim, we selected 438 unenhanced scans from the ADNI-1 dataset, preprocessed them, and injected the preprocessed scans to the ConvLSTM-based neural network for automatic feature extraction and binary/multiclass classification. On test data, CLAUDIA achieved very encouraging results that highlight the superiority of the multiclass classifier in comparison to the two binary classifiers. On the basis of the achieved outcomes, we demonstrated that CLAUDIA, being the first to extend the applicability of a ConvLSTM-based neural network to MCI for effectiveness verification, represents a promising scan-, DL-based decision-support system for the automatic diagnosis of Alzheimer's prodromal stage and disease from 3D brain MR. Moreover, its cloud thus machine-independent nature ensures a full reproducibility of the implementation while guaranteeing cost saving and sustainability.

Heart rate (HR) recording is a non-invasive, cheap and time-efficient tool for continuous cardiovascular monitoring through wearable technologies in sport applications directly on field. Although, HR measures cannot inform equally on all aspects of cardiac responses to training, given the individual HR kinetic that depends on internal and external influencing factors. Knowledge of the training context is required to correctly compute and interpret HR-derived indices. Training context is characterized by the training phases, their distribution and training load. The aim of this study is to develop an algorithm for automatic detection of training phases in HR series to boost signal processing for athletic cardiovascular monitoring with wearable technologies. The algorithm computes the start and end times of the training phases. It exploits the variance of HR series computed over moving overlapping windows to detect automatically training transition phases. The algorithm was tested on HR series acquired during middle distance running and jogging. The algorithm showed promising results: mean errors were globally lower than 5 s and percentage error did not exceed 5%. Thus, the fully automatic algorithm for detection of training phases can boost HR signal processing for reliable computation and interpretation of HR-derived indices during continuous cardiovascular monitoring with wearable sensors in athletes.