Amir Mohammad Amiri

Fog Data: Enhancing Telehealth Big Data Through Fog Computing

The size of multi-modal, heterogeneous data collected through various sensors is growing exponentially. It demands intelligent data reduction, data mining and analytics at edge devices. Data compression can reduce the network bandwidth and transmission power consumed by edge devices. This paper proposes, validates and evaluates Fog Data, a service-oriented architecture for Fog computing. The center piece of the proposed architecture is a low power embedded computer that carries out data mining and data analytics on raw data collected from various wearable sensors used for telehealth applications. The embedded computer collects the sensed data as time series, analyzes it, and finds similar patterns present. Patterns are stored, and unique patterns are transmited. Also, the embedded computer extracts clinically relevant information that is sent to the cloud. A working prototype of the proposed architecture was built and used to carry out case studies on telehealth big data applications. Specifically, our case studies used the data from the sensors worn by patients with either speech motor disorders or cardiovascular problems. We implemented and evaluated both generic and application specific data mining techniques to show orders of magnitude data reduction and hence transmission power savings. Quantitative evaluations were conducted for comparing various data mining techniques and standard data compression techniques. The obtained results showed substantial improvement in system efficiency using the Fog Data architecture.

WearSense: Detecting Autism Stereotypic Behaviors through Smartwatches

Autism is a complex developmental disorder that affects approximately 1 in 68 children (according to the recent survey conducted by the Centers for Disease Control and Prevention—CDC) in the U.S., and has become the fastest growing category of special education. Each student with autism comes with her or his own unique needs and an array of behaviors and habits that can be severe and which interfere with everyday tasks. Autism is associated with intellectual disability, impairments in social skills, and physical health issues such as sleep and abdominal disturbances. We have designed an Internet-of-Things (IoT) framework named WearSense that leverages the sensing capabilities of modern smartwatches to detect stereotypic behaviors in children with autism. In this work, we present a study that used the inbuilt accelerometer of a smartwatch to detect three behaviors, including hand flapping, painting, and sibbing that are commonly observed in children with autism. In this feasibility study, we recruited 14 subjects to record the accelerometer data from the smartwatch worn on the wrist. The processing part extracts 34 different features in each dimension of the three-axis accelerometer, resulting in 102 features. Using and comparing various classification techniques revealed that an ensemble of 40 decision trees has the best accuracy of around 94.6%. This accuracy shows the quality of the data collected from the smartwatch and feature extraction methods used in this study. The recognition of these behaviors by using a smartwatch would be helpful in monitoring individuals with autistic behaviors, since the smartwatch can send the data to the cloud for comprehensive analysis and also to help parents, caregivers, and clinicians make informed decisions.

Emotional reactivity monitoring using electrodermal activity analysis in individuals with suicidal behaviors

Suicide, considered as one of the leading causes of death, has not been given enough attention in order to reduce its rate. The problem addressed in this paper is the analysis of the relation between an extra stimulus and physiological datas responses. In order to record the physiological data set from multiple subjects over many weeks, we used an acoustic startle during a Paced Auditory Serial Addition Task (PASAT) test that spontaneously leads subjects to real emotional reactivity, without any deliberate laboratory setting. Crucially, we show that, by inducing anxiety during the test, changes appear in Electrodermal activity, Electrocardiogram, Heart Rate and Respiration Rate. A wide range of physiological features from various analysis domains, including modeling, time/frequency analysis, an algorithm and etc., is proposed in order to find the best emotional reactivity feature to correlate them with emotional states which can be considered as a suicide factor. More specifically, this paper is focused on the EDA data analysis. Experimental results highlight that all cited techniques perform well and we achieved a high resolution of tonic and phasic components which allow us to measure the latency, onsets and amplitudes of EDA responses to a stimulus. This paper follows the association of recommendations for advancement of health care instruments.

Mobile Phonocardiogram Diagnosis in Newborns Using Support Vector Machine

Phonocardiogram (PCG) monitoring on newborns is one of the most important and challenging tasks in the heart assessment in the early ages of life. In this paper, we present a novel approach for cardiac monitoring applied in PCG data. This basic system coupled with denoising, segmentation, cardiac cycle selection and classification of heart sound can be used widely for a large number of the data. This paper describes the problems and additional advantages of the PCG method including the possibility of recording heart sound at home, removing unwanted noises and data reduction on a mobile device, and an intelligent system to diagnose heart diseases on the cloud server. A wide range of physiological features from various analysis domains, including modeling, time/frequency domain analysis, an algorithm, etc., is proposed in order to extract features which will be considered as inputs for the classifier. In order to record the PCG data set from multiple subjects over one year, an electronic stethoscope was used for collecting data that was connected to a mobile device. In this study, we used different types of classifiers in order to distinguish between healthy and pathological heart sounds, and a comparison on the performances revealed that support vector machine (SVM) provides 92.2% accuracy and AUC = 0.98 in a time of 1.14 seconds for training, on a dataset of 116 samples.

PhonoSys: Mobile Phonocardiography Diagnostic System for Newborns

Heart murmurs have been found to be a life threatening condition for the newborns who are born with cardiac abnormalities. The first sign of pathological changes of heart valves appears in phonocardiogram which contains very useful information about cardiovascular system. It is a challenging venture to distinguish pathological murmurs from innocent ones. In this paper we have developed a diagnostic algorithm called PhonoSys to analyze PCG using random forest. PhonoSys algorithm will run on mobile devices for remote PCG analysis. We recorded PCG signals from 120 newborns who are either healthy or with cardiac abnormalities. Eventually, in this study, 97.6% accuracy, 96.8% sensitivity, and 98.4% specicity were obtained to classify between innocent and pathological murmurs.

m-QRS: An efficient QRS detection algorithm for mobile health applications

When using the available m-health systems, ECG data for a small duration is recorded and sent to a server for processing and arrhythmia detection. Since arrhythmia occurrence is not so frequent in early stages, a need is felt to develop a real time and continuous arrhythmia monitoring system on the phone itself. This paper provides a novel approach to detect QRS complexes from a high fidelity ECG data obtained from B.E.A.T. ® hardware for arrhythmia monitoring in real time. Our approach referred to as m-QRS uses continuous wavelet transform at its kernel and its efficiency is compared to that of Pan-Tompkinss which is a standard QRS detection algorithm widely used for arrhythmia detection. It was found that our algorithm uses lesser computation time when compared to Pan-Tompkins and was found to be mobile friendly. This provides an opportunity to develop further algorithms to perform continuous and real-time arrhythmia monitoring on affordable smartphones without internet dependability.

Hand Motion Detection in fNIRS Neuroimaging Data

As the number of people diagnosed with movement disorders is increasing, it becomes vital to design techniques that allow the better understanding of human brain in naturalistic settings. There are many brain imaging methods such as fMRI, SPECT, and MEG that provide the functional information of the brain. However, these techniques have some limitations including immobility, cost, and motion artifacts. One of the most emerging portable brain scanners available today is functional near-infrared spectroscopy (fNIRS). In this study, we have conducted fNIRS neuroimaging of seven healthy subjects while they were performing wrist tasks such as flipping their hand with the periods of rest (no movement). Different models of support vector machine is applied to these fNIRS neuroimaging data and the results show that we could classify the action and rest periods with the accuracy of over 80% for the fNIRS data of individual participants. Our results are promising and suggest that the presented classification method for fNIRS could further be applied to real-time applications such as brain computer interfacing (BCI), and into the future steps of this research to record brain activity from fNIRS and EEG, and fuse them with the body motion sensors to correlate the activities.

Human motion identification using functional near-infrared spectroscopy and smartwatch

Copious amounts of people around the globe currently suffer from ailments in moving, which range from neurodegenerative diseases to colossal accidents. In this study, oxygenated hemoglobin of the brain is monitored using a functional near infrared spectroscopy coupled with a smart watch to detect kinetic activity. It was seen that as participants flipped their left or right hands, at different speeds, there was a detectable increase in oxygenated hemoglobin in the cerebral motor cortex. These promising results could later be used in the advancement of applications based around telehealthcare and brain-computer interface.

Bivariate autoregressive state-space modeling of psychophysiological time series data

Heart rate (HR) and electrodermal activity (EDA) are often used as physiological measures of psychological arousal in various neuropsychology experiments. In this exploratory study, we analyze HR and EDA data collected from four participants, each with a history of suicidal tendencies, during a cognitive task known as the Paced Auditory Serial Addition Test (PASAT). A central aim of this investigation is to guide future research by assessing heterogeneity in the population of individuals with suicidal tendencies. Using a state-space modeling approach to time series analysis, we evaluate the effect of an exogenous input, i.e., the stimulus presentation rate which was increased systematically during the experimental task. Participants differed in several parameters characterizing the way in which psychological arousal was experienced during the task. Increasing the stimulus presentation rate was associated with an increase in EDA in participants 2 and 4. The effect on HR was positive for participant 2 and negative for participants 3 and 4. We discuss future directions in light of the heterogeneity in the population indicated by these findings.

Automatic sleep apnea detection using fuzzy logic

The obstructive sleep apnea (OSA) is one of the most important sleep disorders characterized by obstruction of the respiratory tract and cessation in respiratory flow level. Currently, apnea diagnosis is mainly based on the Polysomnography (PSG) testing during sleeping hours, however, recording the entire signals during nights is a very costly, time-consuming and difficult task. The goal of this study is to provide and validate an automatic algorithm to analyze four PSG-recordings and detect the occurrence of sleep apnea by noninvasive features. Four PSG signals were extracted from oxygen saturation (SaO2), Transitional air flow (Air Flow), abdominal movements during breathing (Abdomen mov.) and movements of the chest (Thoracic mov.). We describe a fuzzy algorithm to compensate the imprecise information about the range of signal loss, regarding the expert opinions. Signal classification is implemented minute-by-minute and for 30 labeled samples of MIT/BIH data sets (acquired from PhysioNet). The obtained data from 18 apnea subjects (11 males and 7 females, mean age 43 years) were categorized in three output signals of apnea, hypopnea and normal breathing. The proposed algorithm shows proficiency in diagnosing OSA with acceptable sensitivity and specificity, respectively 86% and 87%.