Showing papers presented at "IEEE International Symposium on Medical Measurements and Applications in 2018"

Abnormality Detection in Mammography using Deep Convolutional Neural Networks

Abstract: Breast cancer is the most common cancer in women worldwide. The most common screening technology is mammography. To reduce the cost and workload of radiologists, we propose a computer aided detection approach for classifying and localizing calcifications and masses in mammogram images. To improve on conventional approaches, we apply deep convolutional neural networks (CNN) for automatic feature learning and classifier building. In computer-aided mammography, deep CNN classifiers cannot be trained directly on full mammogram images because of the loss of image details from resizing at input layers. Instead, our classifiers are trained on labelled image patches and then adapted to work on full mammogram images for localizing the abnormalities. State-of-the-art deep convolutional neural networks are compared on their performance of classifying the abnormalities. Experimental results indicate that VGGNet receives the best overall accuracy at 92.53% in classifications. For localizing abnormalities, ResNet is selected for computing class activation maps because it is ready to be deployed without structural change or further training. Our approach demonstrates that deep convolutional neural network classifiers have remarkable localization capabilities despite no supervision on the location of abnormalities is provided.

ECGNet: Deep Network for Arrhythmia Classification

Abstract: Cardiac arrhythmias are presently diagnosed by manual interpretation of Electrocardiography (ECG) signals. Automated ECG interpretation is required to perform efficient screening of arrhythmia from long term ECG data. Existing automated ECG interpretation tools however require extensive preprocessing and knowledge to determine relevant features. Thus there is a need for a comprehensive feature extractor and classifier to analyze ECG signals. In this paper, we propose three robust deep neural network (DNN) architectures to perform feature extraction and classification of a given two second ECG signal. The first network is a Convolutional Neural Network (CNN) with multiple kernel sizes, the second network is a Long Short Term Memory (LSTM) network and the third network is a combination of CNN and LSTM based feature extractor, CLSTM network. The proposed networks are end to end networks which can be directly trained without any preprocessing. The networks were trained and tested with the MITDB ECG dataset on three classes Normal (N), Premature Ventricular Contraction (PVC) and Premature Atrial Contraction (PAC). The best model CLSTM gave an accuracy of 97.6%. Further, transfer learning is showcased on the best performing network for use with multiple ECG datasets requiring training only on the final three layers. The results showcase the potential of the network as feature extractor for ECG datasets. Our results outperform the state-of-the art works on ECG classification on several metrics.

Human Activity Recognition by Wearable Sensors : Comparison of different classifiers for real-time applications

Abstract: In recent years, there is a growing interest in Human Activity Recognition (HAR) systems applied in healthcare. A HAR system is essentially made of a wearable device equipped with a set of sensors (like accelerometers, gyroscopes, magnetometers, heart-rate sensors, etc…) and a classifier able to recognize the activity performed. In this study we focused on the choice of the classifier, since there isn’t a unique and consolidated methodology for HAR. The main aim of the study is to compare the performances of 5 classifiers, based on machine learning. Furthermore, we analyzed advantages and disadvantages of their implementation onto a wearable and realtime HAR system. We acquired magnetic and inertial measurement unit (MIMU) signals from 15 young volunteers. For each subject, we recorded 9 signals from tri-axis accelerometer, gyroscope and magnetometer. All signals were divided in 5s-windows and processed to extract 342 features in time, frequency and time-frequency domains. By means of two feature selection steps (correlation-based and genetic algorithm), we reduced the number of features to 69. These features were used as input for the following 5 classifiers: K-Nearest Neighbor (KNN), Feedforward Neural Network (FNN), Support Vector Machines (SVM), Naive Bayes (NB), and Decision Tree (DT). Our results showed that all classifiers were able to correctly recognize more than 90% of activities. The best performances were obtained by KNN. Analyzing advantages and disadvantages of each classifier for its implementation by means of a microcontroller the most suitable was DT. In fact, this classifier can be easily implemented, it has low memory and computational requirements, and it allows for a further reduction of the required features.

An Improved Approach for Computer-Aided Diagnosis of Breast Cancer in Digital Mammography

Abstract: Breast cancer keeps on being a major medical problem among women around the world. Early detection of breast cancer can expand the treatment options and consequently would increase the surviving possibilities for patients. In this paper, a new Computer Aided Diagnosis (CAD) system is proposed for breast cancer diagnosis in digital mammography. An improved technique for feature extraction based on Wavelet-Based Contourlet Transform (WBCT) is investigated to obtain the features of the Region of Interest (ROI), allowing for accuracy improvement over other standard approaches. Aiming to reduce the features dimensions, we have proposed a hybrid feature selection approach in which the Genetic Algorithm (GA) and the Support Vector Machine (SVM) are combined along with the Mutual Information (MI) in order to select the best combination of tumor indicators, with maximal discriminative ability. The Particle Swarm Optimization (PSO) is also investigated instead of GA for performance evaluation of both methods. The selected features are then submitted to the kernel SVM classifier and its performance is compared with the traditional machine learning classification techniques. The diagnosis accuracy of the implemented CAD system is demonstrated by investigating several experimental datasets and comparing the results with other diagnosis approaches. The results show that the proposed CAD system (WBCT + GA-SVM-MI + kernel SVM) is superior over other techniques in terms of the classification accuracy (97.5% for normal-abnormal and 96% for benign-malignant), while keeping the computational requirements as low as possible.

Measurement system based on RBG camera signal for contactless breathing pattern and respiratory rate monitoring

Abstract: Vital signs monitoring is pivotal in clinical environments and emerging in home-based healthcare applications. Different kind of sensors can be used to monitor the breathing pattern and the respiratory rate. However, respiration rate remains the least measured vital sign in several scenarios, due to the intrusiveness of the sensors usually adopted. In this paper, we propose a measuring system for the low-cost and contactless analysis of respiration. The system analyses a video recorded by commercial RGB camera and extracts respiratory pattern from the changes in the intensity of the light reflected back at the level of the upper chest. Then it allows estimating breath-by-breath respiratory rate from such pattern. The proposed system was tested on six healthy volunteers seated in front of the camera, with both slim-fit and loose-fit clothes. Results show promising performances for the non-intrusive monitoring of both breathing pattern and breath-by-breath respiratory rate over time. Future developments will be devoted to investigating the feasibility of the proposed measurement system for long-term observation of respiratory parameters, for subject monitoring at home or during clinical examinations (i.e., MR imaging).

Convolutional Auto-Encoder Based Deep Feature Learning for Finger-Vein Verification

Abstract: This paper presents a novel deep learning-based method that integrates a Convolutional Auto-Encoder (CAE) with Convolutional Neural Network (CNN) for finger vein verification. The CAE is used to learn the feature codes from finger vein images and the CNN is used to classify finger vein from these learned feature codes. The CAE consists of a finger vein encoder, which extracts high-level feature representation from raw pixels of the images, and a decoder which outputs reconstruct finger vein images from high-level feature code. Experimental study proves that the proposed deep learning based method has superior performance in learning features than traditional method without any prior knowledge, presenting a good potential in the verification of finger vein.

Electrodermal Activity based Classification of Induced Stress in a Controlled Setting

Abstract: It has become common for people to experience stress, mainly because of its eclectic nature – physical, psychological, emotional, social, etc. Unmonitored stress may prove harmful to one’s health resulting in even chronic diseases. Since stress is very subjective, stress management is not straightforward. Many attempts have been made to detect and quantify stress. However, an accurate assessment can be made from physiological measurements only. In this study, we have demonstrated how electrodermal activity (EDA), which represents the sympathetic response to stress, could be used for accurate classification of stress by developing a machine learning based classification model. 30 participants were subjected to Trier Social Stress Test (TSST), and EDA and accelerometer data were recorded using a wrist-worn device. Datasets containing stress and non-stress periods were segmented and manually tagged for model training, based on recorded stress protocol timeline. A kNN-classifier model was trained on datasets from 15 participants and tested on datasets from the remaining 15 participants, and the results were verified with salivary cortisol levels recorded before and after TSST. The proposed kNN classifier has sensitivity and specificity of 94% and 93% respectively. Motion corruptions due to hand movements were detected using the accelerometer data and were classified as ‘motion affected’. The classifier was able to classify – the baseline regions of all participants as non-stress, 93% of the TSST regions as stress and 63% of the post-stress regions as non-stress.

Assessment of cardio-respiratory rates by non-invasive measurement methods in hospitalized preterm neonates

Abstract: Noncontact monitoring of physiological signsin preterm infants is necessary in order to increase neonatal comfort and open new possibilities to home respiration monitoring. The presented solution is obtained by using commercially available digital webcamera (WeC) and Eulerian video magnification (EVM) algorithm which amplify the small changes on the sequence of the recorded images, that would be invisible to the naked eye. With this method, the respiration rate (RR) is extracted as the dominant frequency in the thorax portion of the patient, and the heart rate (HR) by observing minor color changes of the skin. The results obtained are compared with the signal from standard multi-parameter monitor (MM). Data collected on 7 patients demonstrate the feasibility of the proposedmeasurement method. The measure of RR and HR show close correlation with the data from the MM, with a root mean square error of 12.2 for the HR and 7.6 for the RR. This technique can be a valuable non-contact monitoring system for the improvement of clinical care in NICU and may representa potential application in home health monitoring.

A radar system for indoor human localization and breath monitoring

Abstract: This paper presents the study of a radar system for indoor human localization and vital signs monitoring. The article illustrates different solutions including linear frequency-modulated continuous-wave (FMCW) mode and recently proposed hybrid schemes, combining FMCW and continuous-wave (CW) interferometry modes. A new operating modality based on the use of the phase variations of the demodulated FMCW signal is also proposed. The presented schemes have been investigated theoretically and simulations have been performed to analyse and compare different algorithms needed to extract distance and breathing information. A radar system working in the 24 GHz ISM band with a bandwidth of 250 MHz has been realised. The sensor is based on the Infineon BGT24MTR11 integrated circuit. As transmitting and receiving antennas an appropriately designed serial array of patch antennas and commercial horns have been used. A series of experiments to evaluate the ability of the radar to estimate both the position of a target and its small displacements have been carried out. Maximum registered errors are about 6 cm in range measurement and $30~\mu \mathrm {m}$ in small movements. All the obtained results are in good agreement with simulations performed with a MatlabTM code reproducing the experimental scenario.

Comparison of Different Motion Capture Setups for Gait Analysis : Validation of spatio-temporal parameters estimation

Abstract: Inertial measurement units (IMUs) are increasingly used in gait analysis because of their portability, low cost and long-distance measuring. The aim of this study focused on the comparison of 2 different inertial sensors setups and algorithms respect to a reference measure system for the estimation of gait spatio-temporal parameters. (a) A marker stereo photogrammetric system (Optitrack) was used as reference goldstandard. Two IMU setups were adopted for the comparison of different algorithms: (b) a single sensor positioned on the trunk, and (c) 2 sensors positioned on the heels. Three healthy subjects performed gait trials at 3 different self-selected speeds: (α) normal, (β) slower than normal, (γ) higher than normal. Data analysis considered signals that had been registered simultaneously by the three setup instrumentations. Among the IMUs signals, acceleration and angular velocity were considered and used for gait parameters estimation. Signal post-processing was performed through algorithms analyzing signals with respect to the local sensor axes, avoiding cumbersome pre-processing. The absolute errors among the spatio-temporal gait parameters obtained by the 3 setups were evaluated. Overall, IMUs allowed an accurate parameters estimation, at all the performed velocities. A consistent correspondence was confirmed by the comparison of estimated spatio-temporal parameters. However, for most of them, the trunk configuration revealed smaller errors with respect to the heels configuration. This may be explained by a better identification of gait events guaranteed by the trunk IMU algorithm. The results demonstrated the suitability and accuracy of the trunk IMU setup and algorithm for the estimation of spatio-temporal parameters during gait. Besides, the trunk IMU setup requires the use of only one sensor instead of two.

A Flexible Front-End for Wearable Electrochemical Sensing

Abstract: This work presents the design and the realization of a flexible front-end circuitry for electrochemical sensing with wearable devices. The hardware combines readout circuitry for amperometric and Open Circuit Potential (OCP) measurements. The sensing platforms are dedicated to lactate and lithium detection in sweat, hence allowing the monitoring of athletes under physical effort. The wearability of the system is ensured by the flexibility of the electronic substrate, its small dimensions that fit an armband case, and the wireless transmission through a Bluetooth Low Energy (BLE) module. The power consumption of the system has been evaluated to be 200mW, with 3.6V on board power supply.

Ambulatory monitoring of respiratory effort using a clothing-adhered biosensor

Abstract: Accurate ambulatory sensing of patterns of respiratory effort has long eluded biomedical researchers. Though the data would inform clinical decision-making, the lack of sensors suitable for such data collection stymies clinical impact. This study describes an approach to sensing respiratory effort via thoracic and/or abdominal excursion in a form that affords longitudinal and continuous adherence. This approach leverages force-based sensors embedded in a clothing-adhered form factor to reduce user inconveniences. The primary benefit of this approach is user acceptance: it can monitor data longitudinally while addressing impactful user inconvenience issues with existing devices. Compared to a ground truth monitor of respiratory effort, and across both cognitive and physical tasks, the present approach resulted in a relative median error of 6.8% and mean absolute error of 1.8 breaths per min (SD=0.14). Sensor location affected performance, with chest-worn sensors outperforming waist-worn sensors. A more granular analysis of temporal markers of the respiratory cycle showed high agreement with ground truth; end-of-expiration temporal markers exhibited the least precision. The results indicate that this approach can be used to monitor respiratory effort accurately and theoretically with high adherence.

Eulerian Magnification of Multi-Modal RGB-D Video for Heart Rate Estimation

Abstract: Eulerian Video Magnification (EVM) has been shown to be highly effective for non-contact, unobtrusive, and non-invasive patient heart rate (HR) estimation systems. EVM is typically applied to RGB video to amplify minute changes in skin color due to varying blood flow, thereby estimating HR. Previous methods require knowledge of the expected HR to optimize the passband to be amplified via EVM. Furthermore, most EVM methods operating on natural light video often fail in low-light environments. This paper proposes a multi-modal selective passband search approach, utilizing predefined EVM passbands, and the use of intelligent data fusion of the three different modalities provided by the Intel RealSense RGB-D camera. We demonstrate the effectiveness of using the color, depth, and near-infrared streams to obtain a consensus HR estimate under various lighting conditions and subject poses. Results indicate that the fusion of HR estimates acquired from each modality is effective and robust to environmental conditions.

Measuring age-related differences in kinematic postural strategies under yaw perturbation

Abstract: The assessment of age-related postural strategies under external perturbation is of great interest for a better evaluation of the risk of falls in healthy humans. In the large majority of the studies, subjects have been perturbed with antero/posterior translations or rotations in the pitch and roll angles in order to investigate the postural strategies adopted for balance control. However, physiological mechanisms involved in the response to continuous yaw perturbation to maintain stable balance in healthy subjects are still unclear. Ten younger subjects (age: 28±3 years) and ten older adults (age: 61±4 years) were asked to stand on the RotoBiT1D under two visual conditions: (a) eyes opened looking at a fixation point, and (b) eyes closed. The platform, driven by an ad-hoc control software, provided two sinusoidal rotations on the horizontal plane with fixed amplitude (±55°) and two different frequencies (0.2 Hz and 0.3 Hz). Kinematics of head, trunk, pelvis, arms, forearms, thighs and shanks body-segments was gathered using eleven inertial measurement units. Body-segment absolute rotations in the transverse plane were compared to platform absolute rotation after fast Fourier transform. The gain (G) and phase lag (φ) of all body-segments were computed and analyzed as a function of age, visual and frequency conditions. G values were statistically lower in older subjects than in younger subjects for all body-segments attesting prominent stiffness and limited ranges of movement. Regarding φ, our results demonstrated that in older subjects, lower limbs, trunk and pelvis anticipated platform movement probably compensating for postural perturbations. In both groups, G decreased and φ delayed progressively by increasing the frequency of the perturbation while similar results were obtained comparing the two visual conditions. In conclusion, to maintain balance under sinusoidal yaw perturbation, older subjects adopted a different motor control strategy compared with younger subjects mainly implying reduced body-movements and the anticipation of the movement.

Dry Electrode Optimization for Wrist-based Electrodermal Activity Monitoring

Abstract: Recent years of healthcare research has shown an increasing demand and interest in wearable personal healthcare systems. This, when combined with the superior utility of Electrodermal activity (EDA) for diverse applications ranging from market research to human stress and sleep quality analysis to seizure detection have driven significant interest towards optimizing dry electrode designs for EDA monitoring. The conventional wet adhesive Ag/AgCl electrodes, which is the gold standard, is used almost universally and provide excellent signal quality. However, they are less comfortable and not suitable for wearable scenarios that involve continuous long-term monitoring. This study focuses on identifying an optimum dry electrode configuration for monitoring EDA from the wrist. It is hypothesized that parameters like electrode material, interelectrode distance and anatomical location of measurement influence the dry electrode design for EDA detection. Accordingly, stainless steel, silver, brass and gold electrodes were fabricated with geometry and dimensions similar to that of commercially available standard wet electrodes. The fabricated electrodes were further investigated with interelectrode separations of 2 cm and 4 cm, both on the ventral and dorsal surfaces of the wrist at 6 cm distance from the carpus, thereby constituting 16 dry electrode configurations. An experimental protocol that spanned over 10 hours per subject was designed to investigate these configurations systematically and to identify the one that yielded the highest correlation with the gold standard. Also, the stabilization period for these dry electrode configurations were quantified as the time taken by the EDA signal to reach within ± 5% of its final value. The experimental study done on 7 subjects (3 females and 4 males, age: 23.3 ± 2.8 years; mean ± SD) illustrated that silver electrodes worn on the dorsal surface of the wrist with an interelectrode separation of 4 cm performed consistently well on all subjects with an average Pearson correlation coefficient of 0.899 ± 0.036, following an average stabilization time of 27.1 minutes.

Feature Extraction in Sit-to-Stand Task Using M-IMU Sensors and Evaluatiton in Parkinson's Disease

Abstract: This work proposes a broad analysis for the de-tection of the most relevant features for the sit-to-stand task analysis, in Parkinson's disease (PD) patients and healthysubjects (H). A group of sixteen PD patients and thirteen H subjects have been analyzed, using one magneto-inertial sensor, while the physician administers the UPDRS clinical scale. The PD group has been examined before and after thepharmacological therapy (respectively, OFF and ON phase), in order to monitor the different states of the PD, which implies changes in motor control. By calculating the features of this task, it has been possible to choose the most reliable indexes, already used in this task in order to identify differences in the score assigned through sensors. In addition to that, it has also been possible to find differences in the features' values which the clinical scale and the physician cannotidentify. Our study highlights how wearable motion sensors can detect statistically significant differences between OFF/ON phase and H subjects that the clinical evaluation can not. We conclude that our method provides a deep analysis of the sit-to-stand task with only one M-IMU, allowing to check PD patient status,providing a method for home care monitoring.

Health@Home: pilot cases and preliminary results : Integrated residential sensor network to promote the active aging of real users

Abstract: This study is realized within the framework of the Health@Home Italian project. The focus of this paper is to provide a description of the experimentation in a pilot case in Veneto region (Italy). The integrated residential sensor network (composed by a mix of domotic equipment and biomedical devices), which will allow older people to improve their life-style in their houses, is discussed, together with the apartments selection (i.e. the specifications and requirements) and the users’ recruitment. The authors will also introduce the analysis of the expected results, based on measurements of preliminary data and signals in living lab and the first feedback from the 13 recruited users. The results of the preliminary tests are used to improve the architecture, following the user-acceptance and the data collection. In this phase of the study, the possible services have been hypothesized, and this aspect will be investigated after the end of the experimentation phase (end of 2018)

Real-time Neonatal Respiratory Rate Estimation using a Pressure-Sensitive Mat

Abstract: We present an approach for real-time respiratory rate (RR) estimation in a neonatal intensive care unit (NICU) using a pressure-sensitive mat (PSM). Real patient data were collected in an NICU from four sources simultaneously: a PSM placed under the patient, a Draeger patient monitor, a video camera placed directly above the patient, and a custom bedside event annotation application running on a tablet. The PSM data were used to develop an algorithm for estimating the patient’s RR. The results were evaluated against impedance pneumography (IP) based RR measurements from the patient monitor. In comparison to the IP estimates, we achieved a mean absolute error of 4.51 breaths per minute (bpm) for 3 hours of data collected from a single patient. Moreover, we show that our newer approach performs better than the previous frequency-based approach that was developed based on patient simulator data. The estimation accuracy achieved by the new algorithm meets clinical requirements for the determination of RR.

Evaluation of wearable IMU performance for orientation estimation and motion tracking

Abstract: Introducing objective wearable IMU measurements of functional movement quality into clinical assessments may improve accuracy of diagnosis. The goal of the present study was to assess the performance of inexpensive wearable IMUs relative to conventional motion capture equipment during controlled movements that are representative of typical human movement. Thirty-five cycles of spine flexion-extension, lateral bending, and axial twisting were simulated by means of a motorized gimbal at speeds of 20 cycles/min and 40 cycles/min. Differences between cycle-to-cycle maximum angle, minimum angle, and ROM values, as well as correlational analyses within IMUs and between IMUs and motion capture, in all movement directions, were compared. All absolute differences in measurements were 0.99) in all movement directions showing reliability between sensors and measurements. Overall, it was revealed that the sensors perform very well in the primary movement direction and one secondary axis; however, correlation in the third axis is suboptimal for orientation estimation and motion tracking.

Italian students and the Grit-S : A self-report questionnaire for measuring perseverance and passion for long-term goals

Abstract: The short Grit Scale (Grit-S) is a self-report questionnaire which measures trait-level perseverance and passion for long term goals. The current study aimed to validate the Italian Grit-S on university students. It was administered to 127 students. Internal consistency, content validity and discriminant validity of the instrument were assessed. An exploratory factor analysis (VARIMAX) confirmed the 2-factors structure of the scale. Content validity was reached with forward-backward translation, achieving semantic equivalence with the original scale. The total Grit-S displayed good internal consistency overall (alpha=.76). It also showed good discriminant validity, as shown by the significant correlations with the RSA [1] and the COPE-NVI [2]. The Italian version of the Grit-S shows good psychometric properties and appears suitable for evaluating grit in Italian setting.

Effect of different smartphone uses on posture while seating and standing

Abstract: In this study, we investigated possible alterations of neck and trunk posture when using smartphones. Fifteen healthy young subjects were asked to perform four activities with the smartphone (gaming, messaging, web surfing, video watching) in two different postures (stand and seat). Kinematics of both neck and trunk was recorded by an optoelectronic system, and the neck, trunk and cranio-cervical angles were extracted to evaluate the variations with respect to the baseline condition. The results showed statistical differences for neck and trunk angles between the two postures for all the activities with an increase of the values in the seated position. No significant differences were found among the activities except for neck angle in both standing and sitting postures. The results highlight that the significant alteration of the neck and trunk posture shows up when the activities are performed in the seated position. Furthermore, the variation of the neck angle depends on the performed activity; in fact, during gaming the flexion-extension of the neck is greater than during video watching. These results suggest that during smartphone use the posture undergoes some changes that may be a potential risk factor to develop neck pain, musculoskeletal fatigue and disorders.

Improvement of Signal Quality During Recovery of Compressively Sensed ECG Signals

Abstract: This paper investigates the application of the newly proposed Kronecker-based method for the reconstruction of compressively sensed electrocardiogram (ECG) signals. By applying the Kronecker-based method, ECG signal acquisition is done in smaller lengths. Collection of smaller length of ECG signals leads to fewer arithmetic operations during the compression phase. Instead of recovering individually in smaller lengths, recovery is done over several concatenated segments. This newly proposed recovery method improves the quality of the reconstructed signal when compared to the traditional recovery done without concatenation. Two random measurement matrices, namely the Gaussian and the Bernoulli matrices, are considered as sensing matrices in this study and the methodology is evaluated using 10 ECG signals acquired from the MIT arrhythmia database. The random Bernoulli matrix is found to provide better quality of recovered compressed ECG signal even with the traditional compressive recovery techniques. Recovery by the newly proposed Kroneckerbased method results in higher SNR in all the ECG signals when the compression ratio (CR) is 25% or 50% and when the CR is 75%, improvement is observed in majority of the ECG signals. Lower CRs provide better reconstruction than higher CRs. The Kronecker-based recovery method may be useful for wearable ECG devices.

Does Reinforcement Learning outperform PID in the control of FES-induced elbow flex-extension?

Abstract: Functional electrical stimulation (FES) is an effective technology in post-stroke rehabilitation of the upper limbs. Because of the complexity of the system, traditional linear controllers are still far to drive accurate and natural movements. In this work, we apply reinforcement learning (RL) to design a nonlinear controller for an upper limb FES system combined with a passive exoskeleton. RL methods learn by interacting with the environment and, to efficiently use the collected data, we simulated large numbers of experience episodes through artificial neural network (ANN) models of the electrically stimulated arm muscles. The performance of the novel control solution was compared to a PID controller on five healthy subjects during planar reaching tasks. Both controllers correctly drove the arm at the target position, with a mean absolute error < 1°. The RL control significantly outperformed the PID in terms of setting time, position accuracy and smoothness. Future trials are needed to confirm these promising results.

Smart textile based on FBG sensors for breath-by-breath respiratory monitoring: tests on women

Abstract: The use of wearable systems for monitoring vital parameters has gained wide popularity in several medical fields, especially in the respiratory monitoring. The focus of the present study is the experimental assessment of a male-fit smart textile based on twelve fiber Bragg grating sensors for the monitoring of respiratory parameters on eight female volunteers. In particular, breath-by-breath temporal respiratory parameters (i.e., respiratory period, breathing frequency, duration of inspiratory and expiratory phases), and breath-by-breath volume variations (i.e., tidal volume measurements) have been estimated by the sensor’s outputs of the t-shirt. Results show good agreement between measurements carried out by the smart textile and the reference instrument (i.e., motion capture system with passive markers), with a bias of 0.002 s for the respiratory period and of 0.014 breaths·$\mathbf{min}^{\mathbf {-1}}$ for breathing frequency. However, bias found in the comparison of breath-bybreath tidal volumes discourages the use of the present smart textile for volume monitoring in female population. The promising results promote further development of the system to allow continuous monitoring in clinical setting and for tele-monitoring purposes.

Automatic Detection of Premature Ventricular Contraction Based on Photoplethysmography Using Chaotic Features and High Order Statistics

Abstract: This paper deals with the analysis of photoplethysmography (PPG) signals for the recognition of premature ventricular contractions (PVC). PGG is an optical method used to measure blood volume changes in a non-invasive manner. As a diagnostic tool, PPG has recently been considered to evaluate the functioning of the cardiovascular system and identify its related disorders. PPG signals from 22 healthy and unhealthy subjects were used in this work. A number of chaotic and statistical features including Lyapunov exponent, skewness, kurtosis, fuzzy entropy and spectral entropy were extracted from the signals and selective features were identified by principle component analysis (PCA) to be used during data classification. Feature reduction method. k-nearest neighbors (kNN), support vector machine (SVM) and neural network were examined as classification algorithms. Results showed that the highest recognition accuracy of 95% and specificity of 90.4% are obtained by the KNN classifier.

Vocal health assessment by means of Cepstral Peak Prominence Smoothed distribution in continuous speech

Abstract: This work deals with an investigation on the Cepstral Peak Prominence Smoothed (CPPS) as a discriminator of vocal health in continuous speech. Individual CPPS distribution and its descriptive statistics in reading and free speech acquired with a headworn microphone were investigated. Two groups of subjects were involved: 72 dysphonic and 39 control volunteers according to videostroboscopy examinations. The 95th percentile showed the highest diagnostic precision in both the speech materials (Area Under Curve of 0.86), with lower values indicating a pathological status of voice. Similar best thresholds were found for both reading and free speech (18.1 dB and 17.9 dB, respectively), but the identical phonemic contents of the reading task allowed higher sensitivity and specificity to be obtained. The voice self-assessment was also evaluated in the healthy and pathological groups by means of a questionnaire, namely the Italian version of the Voice Activity And Participation Profile. Significantly different scores were obtained by the two groups in all the sections of the questionnaire, thus highlighting that vocal problems are actually perceived by dysphonic people. Moreover, the 95th percentile resulted in a strong correlation with the sections of self-perceived voice problem and daily communication.

Simultaneous Measurement of Heart Sound, Pulse Wave and Respiration with Single Fiber Bragg Grating Sensor

Abstract: Heart sound, pulse wave and respiration are simultaneously measured with single fiber Bragg grating(FBG) sensor. The FBG sensor is inscribed in optical fiber so that it is quite thinner than conventional electrical sensor, even enough to be weaved in clothes. In this study, single FBG sensor is taped near the Tricuspid area and on the carotid artery with the surgical tape. Measured data is denoised through digital Butterworth filter with low pass filter (0.2 Hz), medium frequency band pass filter (0.5 Hz - 5.0 Hz) and higher frequency bandpass filter (35.0 Hz - 55.0 Hz) to extract the information of respiration, pulse wave and heart sound, respectively. The lower frequency signal follows the reference respiration sensor. The medium and higher frequency signal contains the features of typical pulse wave and heart sound waveform. The pulse wave measured on precordium contains the feature of apex cardiogram. The result also indicate that the timing of the heart sound is slightly delayed on the carotid artery. These results indicate the possibility of wearable, continuous vital sigh measurement system.

Deep Convolutional Pixel-wise Labeling for Skin Lesion Image Segmentation

Abstract: Melanoma is one of the deadliest form of cancer with an increasing incidence rate. The development of automatic diagnostic tools for the early detection of skin cancer lesions in dermoscopic images can help to reduce melanoma-induced mortality. In this paper, we present an automatic method for skin lesion image segmentation based on a deep learning algorithm for pixel-wise labeling. Experimental results have been obtained by testing two network architectures on publicly available data and, in order to show that the used approach is not data set related, we have used the ISIC database for training the network and the PH2 database for testing. The results show that the proposed approach achieves a very accurate segmentation even in presence of hair and air/oil bubbles. An additional contribution of this work is the development of a semi-automatic GUI for data annotation that can be used to generate more test images.

Tetrapolar Low-Cost Systems for Thoracic Impedance Plethysmography

Abstract: Bioelectrical impedance analysis applied to the pneumographic investigation is a technique for monitoring the breathing activity through the measurement of variations in the trans-thoracic electrical impedance. In this paper we present, a low-cost re-configurable measurement system. The system is based on a 4-electrode voltamperometric technique and a network of inertial sensors for correction of arms motion artifacts. The trans-thoracic impedance is acquired via an ad hoc programmed LabVIEW software. A correction algorithm, based on the correlation between the acquired signal and the motion artifact, is proposed. A preliminary metrological assessment of the system is performed to evaluate the accuracy and sensitivity of patient breath monitoring. Results show a high accuracy in a 100 Ω range of measurement. The proposed algorithm allows for the estimation of the thoracic impedance with a maximum error of 30% and to neglect the phase shift between the breath and the movement signals.

Investigation of EEG Noise and Artifact Removal by Patch-Based and Kernel Adaptive Filtering Techniques

Abstract: Electroencephalography (EEG) signals are usually corrupted with several unwanted noise and artifact sources, which lead to poor signal quality and wrong clinical diagnosis. This paper investigates, for the first time, a 1-D implementation of the patch-based NonLocal Means (NLM) algorithm, which was used for 2-D image processing, in removing (1) the Additive White Gaussian Noise (AWGN) and (2) artifacts from EEG signals. A critical comparison between the NLM approach and one of the most effective adaptive filtering techniques, Kernel Recursive Least Squares (KRLS), is made for both removing the AWGN noise and artifact sources. This comparison is demonstrated by investigating several EEG datasets using different evaluation metrics such as the Signal Noise Ratio (SNR), Mean Square Error (MSE), cross correlation, and computational time. The results show that the NLM-based approach more effectively reduces the AWGN/colored noise than the KRLS adaptive filtering. The results also reveal that the NLM approach fails to remove the Electrocardiogram (ECG) artifact in an artifacts contaminated EEG signal, while the KRLS adaptive filter significantly removes most of this ECG component. This reveals that the patch-based techniques are efficient and promising in removing the AWGN/colored noise sources but they are less successful in suppressing interference artifacts.