Alumni of Smart Environments
Pervasive Computing Research Centre alumni.
Wellness Mobile Phone – Using technology to assist persons with Alzheimer's disease and their carers
The aim of our research is to identify and alleviate a set of problems associated with Alzheimer's Disease (AD) using services deployed through smartphone technology. In order to achieve this aim a number of smartphone services will be created, tested and evaluated by both a control group and AD patients.
A key focus of our work will be on the comparison of various levels of user support provided while training both AD patients and their carers in technology use.
This work will realise benefits by assisting AD patients and their carers to effectively manage their illness and allow patients to live independently at home for longer.
email address: Armstrong-N1[at]email.ulster.ac.uk
Processing and Visualization of Body Surface Potential Maps
According to the World Health Organization (WHO) cardiovascular disease (CVD) is the number one cause of death worldwide. A diagnosis of many CVD pathologies are assisted with a primary diagnostic tool called the 12-lead electrocardiogram (ECG).
The 12-lead ECG is the most popular non-invasive diagnostic tool for assessing cardiac functionality. It employs 10 electrodes and produces 12 waveforms, also known as leads. These waveforms are examined by a clinician to assist in making a diagnosis. This traditional ECG system does not, however, place electrodes on the right side of the chest or on the patient's back. Therefore, the 12-lead ECG misses important information.
Fortunately, an alternative system, namely the Body Surface Potential Map (BSPM) employs up to 219 electrodes which are placed around the entire torso. This approach has been deemed more accurate for diagnosing certain cardiac pathologies when compared to the 12-lead ECG. Nevertheless, the BSPM has yet to see routine clinical practice for a number of reasons, i.e. the lack of a common BSPM storage format, visualisation techniques, standards and clinical practicalities.
This research seeks to contribute to the former issue, which is to develop a common BSPM storage format and to visualise BSPM information coherently.
email address: Bond-R[at]email.ulster.ac.uk
An investigation into improving the usability of social media for older users and their carers.
The Internet is now so integral to our lives that many of us could now not live without it. One generation who is almost forgotten in this Internet age is the elderly. They are on the rise with an predictions estimating that by 2034 they will comprise 23% of the UK population.
One way to address the needs of the elderly is to connect them socially via the Internet using Web 2.0 Technology. Web 2.0 is made up of various components which are made possible through Rich Interfaces that exist that allow a number of surfaces such as Social Networking or participation of people who are able to share their lives to the world.
Recent figures suggest that only 7% of over 65s have even used a social network. The focus of this work is to design interfaces intuitively using Web 2.0 technology taking into consideration interaction design.
email address: Boyd-K5[at]email.ulster.ac.uk
Gait analysis: statistical and machine learning approach in gait monitoring and analysis
Falls amongst the elderly are becoming a major problem, according to statistics 68% of elderly hospitalization for injury were fall related and the increase in percentage of injuries due to falls rise with ageing. Gait is a pattern to describe the movement of walking.
Walking performance deterioration can be caused by neurodegenerative diseases such as Parkinson's and Alzheimer's disease.
The aim of the research is to explore the correlation of gait patterns with gait related clinical conditions, to develop and apply machine learning (ML) and statistical approaches in supporting gait analysis and assessment. Research will focus on low cost wearable sensor devices such as accelerometers and SMART mobile phones to provide a non-invasive, cost effective solution to monitoring and assess changes to gait of patients whilst allowing them to get along with their everyday activities.
email address: Chan-H[at]email.ulster.ac.uk
Intelligent technologies for Smart garments
External Supervisor: Dr. Roger Armitage (Adidas Wearable Sports Electronics)
Smart garments, capable of recording vital sign, position and kinetic information, have the potential to provide crucial feedback for the substantiation of health and wellness into later life.
These Wearable technologies are becoming increasingly sophisticated. Nevertheless, it is clear that there is still a number of questions that remain unanswered with regards to their desig for the elderly.
The current work, funded by the ESRC under the the UK joint research council's New Dynamics of Ageing Programme, will be focused beyond the core acquisition techniques for recording heart vital signs information and will investigate the design, implementation and evaluation of new technological systems which may be embedded within smart garments.
email address: Cleland-I2[at]email.ulster.ac.uk
Computer-Based Assessment and Diagnosis of Parkinson's Disease
External Advisor: Dr. David Craig (Belfast City Hospital/Queens University Belfast)
Current methods of assessment/diagnosis of Parkinson's Disease (PD) are somewhat subjective due to the complexity of the disease.
This work endeavours to create computer-based assessment/diagnosis tools to help make the process more objective.
Two assessment/diagnosis tools have been created to date.
The second tool aims to differentiate between PD related tremor and essential tremor. Each of the tools been evaluated and results show that both tools can differentiate between PD and non-PD.The first of which focuses on identifying motor-related features of PD such as rigidity, akinesia (a stoppage in movement) and bradykinesia (a slowness of movement).
Classification of Body Surface Potential Maps through Dimensionality Reduction
Cardiac assessment is possible through the analysis of the electrical signals recorded from the body torso, namely the Electrocardiogram (ECG). Current activities in this field are considering the usefulness of the Body Surface Cardiac Map, whereby a large number of recording sites are employed producing somewhere in the region of 80+ ECG lead recordings.
From a diagnostic or classification perspective, different pathologies exhibit different electrical characteristics on the ECG and as such only selected portions of the input feature space of the cardiac map require consideration to produce a diagnostic statement.
This research address the concept of dimensionality reduction i.e. consideration and identification of the most relevant clinical information required to make a diagnosis (from a computational knowledge discovery perspective). In addition to the aforementioned, a framework to process and classify the ECG, based on the reduced information content, into a specific diagnostic class is required.
In the given work, this will be largely based on individual classifiers developed with emerging artificial intelligence techniques and mixture of expert strategies to produce the final collaborative result.
Mark is currently a Lecturer in the Computer Science Research Institute
A Task-based Approach to Work Experience Reuse
Supervisors: Dr. Liming Chen, Dr. David Patterson, Dr. Hui Wang, Dr. Bo Hu (SAP Mentor)
Work experience reuse is becoming a technological challenge in the emerging knowledge society. This project aims to propose a task-based approach to facilitate experience reuse in knowledge-intensive work environments.
This project is funded by SAP Research and it is carried out in the context of EU projects NEPOMUK (FP6) and MATURE (FP7) with the particular focus on Task Management.
The key focuses of our work will be on the development of models and ontologies for task-based experience capturing and representation, semantic similarity algorithms for reusable experience recommendation, and a integrated system for task-based experience reuse.
Data driven selection of optimal electrocardiographic recording sites from body surface potential maps
External Advisor: Prof. R. Lux (University of Utah)
The Electrocardiogram (ECG) is the most widely used test in the detection of cardiac abnormality. It is widely acknowledged, however, that more stands to be gained from the ECG through enhanced acquisition, processing and interpretation.
Based on this, the current research aims to suggest new and improved methods for selecting optimal recording sites in electrocardiography.
This research was realised through the testing of a set of four hypotheses that were formulated based on reviewed literature.
This literature review highlighted a number of limitations in the techniques currently used to select electrocardiographic leads. In particular this related to the rationale for lead selection, and the algorithms employed.
Four related investigations were conducted using a dataset consisting of a set of 192 lead body surface potential maps (BSPMs). In the first investigation, two limited lead systems, consisting of four and five recording sites, were designed using eigenvectors derived from a principal component analysis (PCA) of the BSPMs.
This technique was chosen so that all information in the BSPMs could be considered in the lead selection process.
This is as opposed to techniques used by previous investigators where only 12 lead ECG information was considered in the design. Described as 'eigen leads' the two lead systems proposed in this study, were compared to other limited lead systems.
The eigen leads, were not found to offer information capture in excess of what had been proposed by previous investigators, but were deemed a suitable alternative.
In a second investigation, techniques were developed to assist in the selection of recording sites for limited lead BSPMs. This study focussed on choosing recording sites for both optimal ECG signal content capture, and optimal diagnostic information content capture.
It was found that this technique produced results comparable with other, more complex approaches, reported in the literature. This was evident with a correlation coefficient of 0.963 and RMS voltage error of 17.2µV obtained in this study in comparison to respective values 0.965 and 16.8µV obtained using previously suggested lead systems.
The development of lead sets for limited lead BSPMs, considering both optimal diagnostic and signal content, facilitated a comparison of the locations of recording sites for each approach. It was found that the suggested recording sites for each approach did not bear similarity.
In the final investigation the techniques considered above were used to investigate enhancements to the 12 lead ECG. This study was based on the rationale that two of the six precordial leads, measured in the 12 lead approach, can be accurately synthesised making the channels required to record them redundant.
The two leads that are synthesised can then be repositioned to alternate locations to provide greater information capture. In these experiments a number of pairs of precordial leads were repositioned to alternate sites using the lead selection algorithm.
This was in contrast to the work of previous investigators, where leads were repositioned based on 'a priori' knowledge. It was found that across all pairs of repositioned leads, a recording site below V1 and a recording site on the posterior surface yielded superior performance.
The presented investigations have consolidated and enhanced the work of previous investigators by providing improved tools and algorithms for lead selection in electrocardiography. These investigations have also resulted in the suggestion of a number of lead systems which are suitable for use in clinical practice.
Dewar is currently a Lecturer in the Computer Science Research Institute
Context-Sharing and Mediated Communication for Smart Environments
Supervisors: Kare Synnes & Peter Parnes (Lulea University of Technology), Chris Nugent
The electrocardiogram (ECG) is today's most widely used non-invasive cardiac assessment and diagnostic tool. It offers a waveform representation of electric voltages obtained (via electrodes) at distinct recording sites on the surface of the human skin.
These waveforms or "leads" are used in medical examinations to gain a diagnostic insight of the human heart.
The most common ECG system offers 12 leads and requires the application of 10 electrodes on the patient's skin.
Reduced Lead Systems (RLS) derive these 12 leads from a reduced (smaller than 10) set of electrodes. Multiple linear regression based models are the foundation for the majority of derivation algorithms used in current RLS.
Focus of this research is the development and evaluation of derivation algorithms offering improved (reduced error by lead derivation) performance.
email address: Guldenring-D2[at]email.ulster.ac.uk
A Decision Support System for Chronic Disease Self Management
Chronic diseases change patients' physical and emotional functioning, reduce their Quality of Life (QOL) and impair their working ability.
For the majority of people with chronic disease, self management has been showed to be effective to improve their QOL. This research focuses on developing decision support algorithms to support patients to self-manage their chronic disease.
Three chronic diseases will be studied, namely, chronic pain, stroke and congestive heart failure. Machine learning and data mining techniques will be applied in the study to monitor patient's life style and behaviour changes in order to provide useful information for self management.
email address: Huang-Y3[at]email.ulster.ac.uk
Combining sensors for smart homes using the Dempster-Shafer theory of evidence
Supervisors: : Dr. Yaxin Bi, Prof. Chris Nugent
The aims of the project is to develop a decision-making system for combining multiple sensors to recognize users' activities within the context of uncertain data within Smart Homes based on the Dampster-Shafer (D-S) theory of evidence.
The project will focus on the development of a computational theory of evidential reasoning and its application to multi-sensor integration.
Through identify the uncertainty existing in Smart Homes and incorporate the overlaps between activities we have built an inference model for activity recognition and implemented a simulated experiment in a bathroom setting using the dataset from the MIT lab. Future plans are to design an experiment to collect the data from our own lab and cross validate the algorithm used in simulations.
email address: Liao-J1[at]email.ulster.ac.uk
Machine learning and statistical approaches to analysis of real time sensory data
The deployment of pervasive computing technology and services has enabled the creation of smart environments which assist people in their daily activities. However, the deployment of technology and services into these smart environments has been obstructed due to many factors, most notably from the lack of proper models, rules and services.
This thesis presents novel solutions for context-sharing intended to make utilisation of smart environments less complex and more cost efficient. It also presents solutions for mediated communication services which uses the power of smart environments to improve quality, flexibility and usability. The proposed solutions have all been evaluated through implementation and testing of proof-of-concept prototypes.
Smart environments can provide a range of different services, such as giving warnings or reminders. Nevertheless, creation and personalisation of services can be complex and time consuming tasks, which promotes the creation of a common approach for tackling the heterogeneous nature of services and data. This thesis proposes such a common approach in the form of the openHome Suite which handles data collection, data analysis, and decision support for smart environments using XML. HomeML and HomeRuleML, two core components of the openHome Suite, supports sharing of context and rules with other research organizations and developers. It simplifies deployment through the creation of models which can be reused between smart environments.
Tools for mediated communication are common in many smart environments. This thesis proposes how to utilise the smart environment to improve multimedia communication. A smart environment enables a user to transfer media content between devices for better quality, cost, and privacy.
This thesis proposes context-aware communication and demonstrates how this can be established automatically by using the developed HomeRuleML approach, an approach which enables the representation and exchange of decision support rules. The study of context-aware communication and automatic establishment of communication has resulted in a new concept for communication called "dynamic groups" which is a simple and powerful group communication service. The concept has been proven through the development and evaluation of a prototype for mobile devices using near field communication.
The research solutions presented in this thesis have been evaluated in the area of remote healthcare, for which simpler deployment of services into smart environments and better communication possibilities for patients becomes possible.
This has the potential knock-on effect of decreased healthcare costs and improved support for the daily activities to improve a patient's wellbeing. Overall the work is done with an overarching aim for smart environments to help people stay out of hospitals in favour of living a richer life in their own homes.
Robust PQRST Complex Detection in ECG Signals
The first stage in computerised processing of the ECG is Beat Detection. This involves recognition of each wave and identifying its beginning and end. The accuracy of beat detection has significant impact on the overall classification process.
The aim of this project is in the improvement of marker insertion in electrocardiogram beat detection.
Methods to be adopted include Non-syntactic best detection, correlation based beat detetcion along with wavelet based analysis and neural networks.
Computational Neural Models for Body Surface Cardiac Data Analysis
Supervisors: Prof. Norman Black, Prof. Chris Nugent
Identifying the best electrocardiographic leads for diagnosing Myocardial Infarction (MI) is a central goal in Body Surface Cardiac Mapping (BSCM) data analysis. To assist with this problem, an artificial intelligence approach is proposed which identifies a subset of leads and processes the information through an enhanced classification framework to provide MI detection.
The proposed model consists of a novel pre-processing stage based on Lead Performance Maps (LPM) and a processing framework based on a Parallel Consensual Neural Network (PCNN) architecture. The LPMs provide a novel approach of identifying the prediction power of every lead, from the PCNN's point of view, and following ranking, provide the rationale for lead selection.
The model was validated using 77-lead (anterior and posterior) BSCMs from 67 patients with MI, and 170 with non-MI. Three features per lead (STT0, STT60 and QRS isointegral) were used as inputs to the model. Following evaluation of the LPMs, 10 leads from the original BSCM were selected based on their prediction power. These leads were subsequently analysed by the PCNN producing sensitivity, specificity and accuracy figures of 85%, 92.16% and 90.14% respectively.
As a benchmark a decision tree approach was tested with the same dataset, however, all 77 leads were used as inputs. Results obtained for sensitivity, specificity and accuracy were 85%, 90.2% and 88.7% respectively. The proposed approach has demonstrated the ability to assist with the problem of identifying the most appropriate leads from those available and through a PCNN provide MI diagnosis with no loss of generalisation.
In comparison with the benchmark, the proposed model provided superior specificity and accuracy figures. Finally, LPMs can be considered as a valuable tool for providing a better understanding of lead-positioning on MI prognosis.
Defining approaches and tools for the creation of shareable datasets for smart home environments
An Agent-based Approach to Decision Support in Complex Environments
Supervisors: Dr. L Chen, Dr. H Wang, Mr. R Sterritt
Pervasive computing and ambient intelligence have recently emerged as one mainstream approach to facilitating working and living activities by providing context aware personalized decision support in dynamic complex environments. To make use of agent based computing for complex system design and implementation, the project aims to develop an agent-based approach to decision support for ambient assisted living.
The approach will particularly focus on novel methods for reusable, context-aware and adaptive activity modelling and knowledge representation mechanisms to support complex reasoning for activity recognition and timely activity assistance. Application scenarios from assistive living, e.g., Smart Homes, would be used for testing and evaluation of the proposed research outcomes.
email address: Okeyo-G[at]email.ulster.ac.uk
Sensor Distribution Optimisation in Smart Environments
Sensors are fundamental in detecting significant events within smart environments as data from sensors determine the appropriateness of subsequent actuator actions.
At present two distribution strategies establish spatial-frequency sensor positions: total coverage of entire sensing area or individualistic sensor positioning based on subjective human decision making. Both methodologies are not data driven deployment strategies.
This project aims to create a novel sensor distribution paradigm by correlating sensor positions with the unique movement behaviour of the inhabitant, specific to the physical characteristics of the environment.
Social Computing & Assistive Living
Social networking and social software are two phenomena representing the latest development on the Web, extensively and successfully used for improving people's social life, and personal experience sharing.
The ageing population often suffers from loneliness or social separation from mainstream society once they retire. Social computing, notably social networking have recently been used for promoting interaction and experience sharing successfully, which can play a vital role to enhance the quality of life.
A core aspect to be examined is how computing technology can make social networking services more pro-active to counter the inherent desire in older people not to interact to the same degree as young people. This project aims to study the factors from a social perspective on how social computing exerts influence on ageing and inclusion.
email address: Quinn-D15[at]email.ulster.ac.uk
Visualisation of Sensor Data for Monitoring People within Intelligent Environments
The lowering cost and increased availability of sensor technology has resulted in the use of more densely populated sensor networks within intelligent environments. This results in far more data being produced, an amount which is difficult for care providers to assess in its raw form.
This project aims to use novel visualisation techniques in order to display information gathered within intelligent environments, allowing stakeholders to perform rapid analysis and diagnosis of the cognitive or physical condition of the inhabitants of an intelligent environment.
email address: Synnott-J2[at]email.ulster.ac.uk
Computer Vision Techniques in Healthcare and Sensorised Environments
Chronic diseases change patients' physical and emotional functioning, reduce their Quality of Life (QOL) and impair their working ability. For the majority of people with chronic disease, self management has been showed to be effective to improve their QOL.
This research focuses on developing decision support algorithms to support patients to self-manage their chronic disease.
Three chronic diseases will be studied, namely, chronic pain, stroke and congestive heart failure. Machine learning and data mining techniques will be applied in the study to monitor patient's life style and behaviour changes in order to provide useful information for self management.
Computational Study of TdP Arrhythmia Caused by LQT2 Gene Defects
Supervisors: Prof. Werner Dubitzky, Prof. Chris Nugent
External Advisor: Antje Krause
Over the past decade, there has been a surge of interest in using wearable sensors in the area of activity recognition and motion function test. Gait analysis is critical for motion function monitoring and assessment. This project aims to explore the use of wearable sensors in gait pattern analysis. The objectives of the research are to investigate what features can be obtained from wearable sensory data; how these features can be used to assess gait patterns, conditions, or treatment outcomes.
A set of feature construction and selection algorithms have been proposed, implemented and tested. A machine learning and statistics platform has been developed to analyse gait features derived from accelerometer data.
The platform has been applied to analysing three types of datasets, i.e. neuro-degenerative disease gait data collected by pair of footswitches; complex region pain syndrome gait data collected by a standalone accelerometer; gait data collected by an accelerometer embedded in a smart mobile phone.
Machine learning techniques have been utilized to differentiate neuro-degenerative diseases and understand condition stages. Results show that wearable sensors can provide a low cost solution for gait monitoring and assessment in telecare.
email address: Yang-M[at]email.ulster.ac.uk
Context-aware Computing for Activity-based Reminders in an Assisted Environment
This research aims to utilize a pervasive computing solution in the form of a sensorised environment and a smart phone (or an 'intelligent messenger'), referred to as iMessenger. Scientific advances are required in the development of new algorithms for the detection of daily activity and the production of activity-aware reminders.
iMessenger collects movement data related to human daily activities using a smart phone with embedded accelerometers, and uses this information to classify activities as 'normal' or 'abnormal' according to a suggested activity schedule. It also issues appropriate feedback via the smart phone.
Towards the realisation of mHealth solutions to support Autism
Supervisor: Dr Mark Donnelly
A novel Health solution was developed. This solution aims to provide effective in-home support to parents of children with autism.
Children with autism often display a range of challenging behaviours, which can only be understood if accurate information regarding the cause and effects is available. A smart phone based application is presented, which aims to support parents by providing a tool that allows them to quickly record and store relevant and immediate information regarding the display of particular behaviour.
This information can then be shared, via an online web portal, with behavioural specialists who are able to assess the collected data and formulate personalised intervention plans that aim to reduce these behaviours.
Lifestyle Support through Efficient ECG Acquisition and Analysis
Supervisor: Prof. Chris Nugent
People are becoming more aware of their health and the lifestyles they live. As people use computers more and more in a wide variety of occupations, they may find it desirable to monitor their lifestyle without having to leave their desk.
Through the analysis of the electrocardiogram (ECG), it is possible to obtain the heartbeat and diagnose a number of conditions associated with abnormal heart function. To obtain an accurate heartbeat, the ECG signals have to be analysed and calculations performed to find occurrences of the QRS complex, a major component of the ECG signal.
The university has designed, and is implementing, an advanced innovative computer peripheral that allows ECG signals to be recorded form the user's forearms as they operate a computer (filed under UK Patent Application Number 0307575.1)
This project entailed creating the software required to interpret the ECG signals so that a user of the peripheral could effortlessly monitor their lifestyle. The system needed to accurately perform ECG beat detection in the background of general computer usage, while providing the user with intermittent notifications of their heart rate. A method of viewing live and historical recordings graphically was also required.
Steven currently works as a Research Associate on the Nestling Project within the Smart Environments Research Group.
Computerised Analysis of Respiratory Sounds
Supervisors: Jim Murray (QUB), Prof. Chris Nugent, Dr. Frank Owens
Frequent measurement of a patient's vitals is a fundamental factor in the care of the critically ill. Physicians use this information constantly in order to make informed decisions on medical procedures and treatments. Continuously monitoring lung sounds allows the physician to observe evidence of pulmonary infection and other cardio-respiratory complications.
Breath sound monitoring has the potential to serve as a clinically useful non-invasive screening tool. The conventional stethoscope approach however has limitations. The information obtained is subjective and is dependant on the physician's own ability to differentiate between sound patterns. In addition to this the human ear is not very sensitive to lower frequency bands.
Therefore the only reliable and quantitative method for an objective assessment of lung sounds is by digital recording followed by subsequent analysis.
Analysis and development of a Bounded Model Checker for on line recognition of Activities Daily Living.
Supervisors: Prof. Enrico Vicario (University of Florence), Prof. Chris Nugent
In our current work we have investigated the potential of a model checking approach for recognition of ADLs within intelligent environments. Model checking is an automatic technique used for verifying finite state systems. In our work we formalize the intelligent environment as a model and with the model checker and we verify the properties of the model.
Each property of the model can be viewed as the composition of the information gleaned from different sensors, each of which subsequently represents an activity.
We have developed a Bounded Model Checker based on a Linear Temporal Logic for the online analysis of a stream of events for the recognition of ADLs. As an instrumental step we also propose a LT logic that combines past operators with real time constraints. To analyse the performance of our approach we have executed a series of several experiments reproducing real settings in a smart environment.
During the experiments we focused on the specific use case of when a user begins an activity and does not complete it within a specified period of time. This is a common feature of persons suffering from dementia and can lead to a number of potential hazards.
To evaluate our approach in a real environment we have also considered the incomplete and inaccurate nature of the information. To manage these kinds if information we have analysed two possible solution: the introduction of a measure of confidence of the extracted information in a Linear Temporal Logic (LTL) and the development of a Probabilistic Model Checker based on Markovian chains.