Skip to main content
main-content

Über dieses Buch

This volume presents the proceedings of the 3rd ICBHI which took place in Thessaloniki on 18-21 November, 2017.The area of biomedical and health informatics is exploding at all scales. The developments in the areas of medical devices, eHealth and personalized health as enabling factors for the evolution of precision medicine are quickly developing and demand the development of new scaling tools, integration frameworks and methodologies.

Inhaltsverzeichnis

Frontmatter

Big Data Analytics for Precision Medicine

Frontmatter

Hybrid Hierarchical Clustering Algorithm Used for Large Datasets: A Pilot Study on Long-Term Sleep Data

Clustering is a popular analysis technique in a modern science full of unlabeled data, hidden dependencies and relations between elements in datasets. The presented study proposes a new hybrid hierarchical clustering method suitable for large datasets. It is based on the combination of effective simple methods. The proposed method was tested and compared with a widely used agglomerative clustering method. Two groups of datasets were used for testing. The first group contains data delivered from real biomedical data and related to a real problem of indication of sleep stages. The second group consists of artificially generated large data. Time, memory consumption, and mutual information were compared.

V. Gerla, M. Murgas, A. Mladek, E. Saifutdinova, M. Macas, L. Lhotska

Epileptic Seizure Prediction with Stacked Auto-encoders: Lessons from the Evaluation on a Large and Collaborative Database

The seizure prediction performance of algorithms based in stacked auto-encoders deep-learning technique has been evaluated. The study is established on long-term electroencephalography (EEG) recordings of 103 patients suffering from drug-resistant epilepsy. The proposed patient-specific methodology consists of feature extraction, classification by machine learning techniques, post-classification alarm generation, and performance evaluation using long-term recordings in a quasi-prospective way. Multiple quantitative features were extracted from EEG recordings. The classifiers were trained to discriminate preictal and non-preictal states. The first part of the feature time series was considered for training, a second part for selection of the “optimal” predictors of each patient, while the remaining data was used for prospective out-of-sample validation. The performance was assessed based on sensitivity and false prediction rate per hour (FPR/h). The prediction performance was statistically evaluated using an analytical random predictor. The validation data consisted of approximately 1664 h of interictal data and 151 seizures, for the invasive patients, and approximately 4446 h of interictal data and 406 seizures for the scalp patients. For the patients with intracranial electrodes 18% of the seizures were correctly predicted (27), leading to an average sensitivity of 16.05% and average FPR/h of 0.27/h. For the patients with scalp electrodes 20.69% of the seizures (84) on the validation set were correctly predicted, leading to an average sensitivity of 17.49% and an average FPR/h of 0.88/h. The observed performances were considered statistically significant for 4/19 invasive patients (≈ 21%) and for 5/84 scalp patients (≈ 6%). The observed results evidence the fact that, when applied in realistic conditions, the auto-encoder based classifier shows limited performance for a larger number of patients. However, the results obtained for some patients point that, in some specific situations seizure prediction is possible, providing a “proof-of-principle” of the feasibility of a prospective alarming system.

R. Barata, B. Ribeiro, A. Dourado, C. A. Teixeira

Deep Learning Techniques on Sparsely Sampled Multichannel Data—Identify Deterioration in ICU Patients

The focus of this paper is to recognize periods of time deviating from the norm using sparsely sampled multichannel signals. The case in question being the ICU, our domain of interest is patient deterioration. In many cases the recording and analyzing of frequently sampled streaming data that can carry more information is not always an option, while at the same time the availability for data recorded at large time intervals is a common occurrence. To address this issue, we examine whether Deep-learning methods can provide efficient results regarding the recognition of different states during the hospitalization, by utilizing hourly multichannel physiological recordings.

A. Chytas, K. Vaporidi, Y. Surlatzis, D. Georgopoulos, N. Maglaveras, I. Chouvarda

Convolutional Neural Networks for Early Seizure Alert System

A general framework of a system for early seizure detection and alert is presented. Many studies have shown high potential of electroencephalograms (EEG) when there are used together with machine learning algorithms for seizure/non-seizure classification task. In this paper, mainly guidelines will be presented on how to use convolutional neural networks for the purpose of highly accurate classification of non-invasive EEG for patients with epilepsy. Convolutional neural networks can be pre-trained on a sample data as described in this paper and then implemented into an application or a device, which readjusts its parameters according to the patient-specific EEG patterns and thus can be further used as a seizure monitoring and alert system. The paper also demonstrated how transfer learning can be applied to create a patient-specific classifier with high accuracy.

T. Iešmantas, R. Alzbutas

Prediction of Cardiac Arrest in Intensive Care Patients Through Machine Learning

Cardiac arrest is a critical health condition characterized by absence of traceable heart rate, patient’s loss of consciousness as well as apnea, with inhospital mortality of ~80%. Accurate estimation of patients at high risk is crucial to improve not only the survival rate, but also the quality of life as patients who survived from cardiac arrest have severe neurological effects. Existing research has focused on demonstrating static risk scores without taking account patient’s physiological condition. In this study, we are implementing an integrated model of sequential contrast patterns using Multichannel Hidden Markov Model. These models can capture relations between exposure and control group and offer high specificity results, with an average sensitivity of 78%, and have the ability to identify patients in high risk.

E. Akrivos, V. Papaioannou, N. Maglaveras, I. Chouvarda

Scientific Challenge—Lung Sounds Analysis

Frontmatter

Α Respiratory Sound Database for the Development of Automated Classification

The automatic analysis of respiratory sounds has been a field of great research interest during the last decades. Automated classification of respiratory sounds has the potential to detect abnormalities in the early stages of a respiratory dysfunction and thus enhance the effectiveness of decision making. However, the existence of a publically available large database, in which new algorithms can be implemented, evaluated, and compared, is still lacking and is vital for further developments in the field. In the context of the International Conference on Biomedical and Health Informatics (ICBHI), the first scientific challenge was organized with the main goal of developing algorithms able to characterize respiratory sound recordings derived from clinical and non-clinical environments. The database was created by two research teams in Portugal and in Greece, and it includes 920 recordings acquired from 126 subjects. A total of 6898 respiration cycles were recorded. The cycles were annotated by respiratory experts as including crackles, wheezes, a combination of them, or no adventitious respiratory sounds. The recordings were collected using heterogeneous equipment and their duration ranged from 10 to 90 s. The chest locations from which the recordings were acquired was also provided. Noise levels in some respiration cycles were high, which simulated real life conditions and made the classification process more challenging.

B. M. Rocha, D. Filos, L. Mendes, I. Vogiatzis, E. Perantoni, E. Kaimakamis, P. Natsiavas, A. Oliveira, C. Jácome, A. Marques, R. P. Paiva, I. Chouvarda, P. Carvalho, N. Maglaveras

Hidden Markov Model Based Respiratory Sound Classification

This paper presents a method based on hidden Markov models in combination with Gaussian mixture models for classification of respiratory sounds into normal, wheeze and crackle classes. Input features are mel-frequency cepstral coefficients extracted in the range between 50 Hz and 2000 Hz in combination with their first derivatives. The audio files are preprocessed to remove noise using spectral subtraction. Our best score achieved in the official ICHBI Challenge second evaluation phase is 39.56.

N. Jakovljević, T. Lončar-Turukalo

An Automated Lung Sound Preprocessing and Classification System Based OnSpectral Analysis Methods

In this work, respiratory sounds are classified into four classes in the presence of various noises (talking, coughing, motion artefacts, heart and intestinal sounds) using support vector machine classifier with radial basis function kernel. The four classes can be listed as normal, wheeze, crackle and crackle plus wheeze. Crackle and wheeze adventitious sounds have opposite behavior in the time-frequency domain. In order to better represent and resolve the discriminative characteristics of adventitious sounds, non-linear novel spectral feature extraction algorithms are proposed to be employed in four class classification problem. The proposed algorithm, which has achieved 49.86% accuracy on a very challenging and rich dataset, is a promising tool to be used as preprocessor in lung disease decision support systems.

Gorkem Serbes, Sezer Ulukaya, Yasemin P. Kahya

Detection of Cough and Adventitious Respiratory Sounds in Audio Recordings by Internal Sound Analysis

We present a multi-feature approach to the detection of cough and adventitious respiratory sounds. After the removal of near-silent segments, a vector of event boundaries is obtained and a proposed set of 126 features is extracted for each event. Evaluation was performed on a data set comprised of internal audio recordings from 18 patients. The best performance (F-measure = 0.69 ± 0.03; specificity = 0.90 ± 0.01) was achieved when merging wheezes and crackles into a single class of adventitious respiratory sounds.

B. M. Rocha, L. Mendes, I. Chouvarda, P. Carvalho, R. P. Paiva

eHealth Systems, Services and Cloud Computing

Frontmatter

A Content-Aware Analytics Framework for Open Health Data

The vision of personalized medicine has led to an unprecedented demand for acquiring, managing and exploiting health related information, which in turn has led to the development of many e-Health systems and applications. However, despite this increasing trend only a limited set of information is currently being exploited for analysis and this has become a major obstacle towards the advancement of personalized medicine. To this direction, this paper presents the design and implementation of a content aware health data-analytics framework. The framework enables first the seamless integration of the available data and their efficient management through big data management systems and staging environments. Then the integrated information is further anonymized at run-time and accessed by the data analysis algorithms in order to provide appropriate statistical information, feature selection correlation and clustering analysis.

L. Koumakis, H. Kondylakis, D. G. Katehakis, G. Iatraki, P. Argyropaidas, M. Hatzimina, K. Marias

Towards Harmonized Data Processing in SMBG

Self-monitoring of blood glucose (SMBG) is the key activity in diabetes management. Patients are required to take measurements and act accordingly, while the physicians use measured data to adjust the therapy. Though the accuracy of individual glucose meters used for SMBG is limited, the main difficulty in interpretation of the recorded data is due to inaccuracy of the records made by patients themselves into the paper diabetic diary. Oftentimes, patients do not record data properly and therefore the data is not reliable for use in determining long-term changes and trends or to use it for further analysis. Therefore, analysis and decision making should rely on the values recorded and stored in glucose memory. The large variety of glucometer models on the market introduce a large problem in using the recorded values since companies which produce and sale glucometers do not necessarily base their data transmission code on accepted standards but they embed custom made code. Data from 37 models of glucometers is transferred into a cloud based platform using previously developed system and available for immediate analysis and for saving into an appropriate health registry in a harmonized structure despite differences in protocols and data structure of different meters. Immediate statistics are given to the physicians upon patient’s checkup. However, general statistical metrics usually do not include metrics on glucose variability, which is one of the most important measurements of glycemic control. We added glycemic variability metrics, including other metrics into tool for data analysis using MATLAB. The output of the analysis can be stored in the system and can be combined with the existing healthcare registries to develop multidimensional analysis for new knowledge discovery. This paper describes the system for acquisition of SMBG data, MATLAB analysis software and the notes on the analysis of the previously discussed data set.

Sara Zulj, Goran Seketa, Ratko Magjarevic

Notarization of Knowledge Retrieval from Biomedical Repositories Using Blockchain Technology

Biomedical research and clinical decision depend increasingly on a number of authoritative databases, mostly public and continually enriched via peer scientific contributions. Given the dynamic nature of data and their usage in the sensitive domain of biomedical science, it is important to ensure retrieved data integrity and non-repudiation, that is, ensure that retrieved data cannot be modified after retrieval and that the database cannot validly deny that the particular data has been provided as a result of a specific query. In this paper, we propose the use of blockchain technology in combination with digital signatures to create smart digital contracts to seal the query and the respective results each time a third-party requests evidence from a reference biomedical database. The feasibility of the proposed approach is demonstrated using a real blockchain infrastructure and a publicly available medical risk factor reference repository.

P. Mytis-Gkometh, G. Drosatos, P. S. Efraimidis, E. Kaldoudi

Gap Analysis for Information Security in Interoperable Solutions at a Systemic Level: The KONFIDO Approach

In this paper, we present a gap analysis study focusing on interoperability of eHealth systems and services coupled with cybersecurity aspects. The study has been conducted in the scope of the KONFIDO EU-funded project, which leverages existing security tools and procedures as well as novel approaches and cutting-edge technology, such as homomorphic encryption and blockchains, in order to create a scalable and holistic paradigm for secure inner and cross-border exchange, storage and overall handling of healthcare data in compliance with legal and ethical norms. The gap analysis relied on desk research, expert opinions and interviews across four thematic areas, namely, eHealth interoperability frameworks, eHealth security software frameworks, end-user perspectives across diverse settings in KONFIDO pilot countries, as well as national cybersecurity strategies and reference reports. A standards-based template has been created as a baseline through which the analysis subjects have been analyzed. The gap analysis identified barriers and constraints as well as open issues and challenges for information security in interoperable solutions at a systemic level. Recommendations derived from the gap analysis will be brought into the forthcoming phases of KONFIDO to shape its technical solutions accordingly.

J. Rasmussen, P. Natsiavas, K. Votis, K. Moschou, P. Campegiani, L. Coppolino, I. Cano, D. Marí, G. Faiella, O. Stan, O. Abdelrahman, M. Nalin, I. Baroni, M. Voss-Knude, V. A. Vella, E. Grivas, C. Mesaritakis, J. Dumortier, J. Petersen, D. Tzovaras, L. Romano, I. Komnios, V. Koutkias

Identification of Barriers and Facilitators for eHealth Acceptance: The KONFIDO Study

In this paper, we present one of the key KONFIDO project’s activities, the identification of key barriers and facilitators regarding eHealth solutions acceptance, focusing on security and interoperability. The methodology presented includes an end-user survey and an end-user workshop, engaging various stakeholders from Europe, in order to gain value out of their experience and insight in real-world healthcare settings. The analysis of the results provides a list of explicitly identified barriers and facilitators of adopting eHealth solutions in a Europe-wide scale, useful in the context of KONFIDO and beyond.

P. Natsiavas, C. Kakalou, K. Votis, D. Tzovaras, N. Maglaveras, I. Komnios, V. Koutkias

A Personalized Cloud-Based Platform for AAL Support to Cognitively Impaired Elderly People

Population ageing due to declining fertility rates and/or rising life expectancy is poised to significantly transform our societies in the upcoming years. Although a lot of work has been done in the field of AAL for the creation of ICT solutions that will prolong and support the autonomous living of elderly individuals with cognitive impairments, these solutions are unable to meet all user needs and/or to provide an easy to use mechanism for further extension with new services. The present paper presents a cloud-based solution that provides easy, transparent, personalized and contextualized access to all the supported AAL services to the cognitively impaired elderly end-users and their caregivers by also offering a mechanism for easy registration and integration of new AAL services into the platform.

Stefanos Stavrotheodoros, Nikolaos Kaklanis, Dimitrios Tzovaras

Enhanced Healthcare System Based on Mobile Communication

This paper aims to develop an enhanced healthcare system that can not only increase the rate of medication adherence through multimedia technology, but also improve the efficacy of diagnosis and treatment. The system comprises five parts including a medicine bag, a home server, a hospital server, a prescription terminal, and a pharmacy sever. The wireless communication technology performs data communication among different parts in the system, which achieves remote monitoring and rapid efficacy assessment. The presented system is intended for patients, physicians, and caregivers, to enable not only medication prescribing and treatment efficacy reporting during illness, but also complete health record collection before illness. The experimental results show that the proposed system can effectively achieve a better healthcare and fast recovery from illness.

Cheng-Huei Yang, Tsung-Che Wu, Hsiu-Chen Huang

Experience of Using the WELCOME Remote Monitoring System on Patients with COPD and Comorbidities

The WELCOME system is an innovative telemonitoring system designed to provide constant monitoring of COPD patients also suffering from other major comorbidities. It consists of a sensors vest capable of recording various vital parameters in real time and transmitting them to the cloud via a tablet PC and wireless connection. In addition, peripheral devices record extra physiological data which is coupled with responses to validated health questionnaires that the patient responds to via the tablet. A dedicated medical decision support system (DSS) and a medical professional user interface support the system in providing automated detection of abnormal conditions. The obtained signals included Heart and Respiratory Rate, Body Posture, SpO2, multi-lead ECG, Auscultation and Electric Impedance Tomography. The system is tested with pilot studies in two European countries, Greece and UK. The preliminary results from the Greek pilot study are presented in this paper, highlighting the main findings from the first operational use of the WELCOME infrastructure.

E. Kaimakamis, E. Perantoni, E. Serasli, V. Kilintzis, I. Chouvarda, R. Kayyali, S. Nabhani-Gebara, J. Chang, R. Siva, R. Hibbert, N. Philips, D. Karamitros, A. Raptopoulos, I. Frerichs, J. Wacker, N. Maglaveras

Biosignals and Biomarkers

Frontmatter

Adipose Tissue as a Biomarker in Data Mining Predictive Models of Metabolic Pathophysiologies

It is well known that the metabolic syndrome emerges as one of the major public health issues worldwide. In diabetes and other metabolism related diseases, further complexity is added in diagnosis and prognosis due to the presence of metabolic syndrome, including obesity. Obesity, which is defined as an excess of body fat, can be described as an underlying risk factor of almost any of the aforementioned metabolic related pathologies. Moreover, a very likely potential link between such pathologies and obesity is the adipose tissue, which functions as an endocrine organ. Since obesity serves as general key to metabolism related disorders and complications, the adipose tissue can be a useful tool in predicting such pathologies. In the present mini review work, several representative studies are discussed with respect to the effectiveness of adipose tissue as a valuable biomarker along with other factors taken into consideration with data mining approaches. Taken together, adipose tissue can be used in data mining as a predictive tool in diabetes, mortality, cardiometabolic risk and other metabolism related pathologies.

O. Tsave, I. Kavakiotis, I. Vlahavas, A. Salifoglou

Portable Near-Infrared Spectroscopy for Detecting Peripheral Arterial Occlusion

The prevalence rate of peripheral arterial disease has been increasing in Taiwan in recent years. The occlusion of artery will reflect on the oxygen saturation of muscle, especially after exercise. A wireless tissue oximetry based on near-infrared spectroscopy is developed in this study to help to detect arterial occlusion. The measurement of change in hemoglobin oxygen saturation was measured on gastrocnemius muscle using air cuff with different pressure to simulate the blockage of blood flow.

W.-C. Lu, S.-H. Lu, M.-F. Chen, T.-C. Fu, K.-P. Lin, C.-L. Tsai

Physiological Monitoring of Cold-Air Stimulated Rhinitis

Patients with vasomotor rhinitis accounts for most of the non-allergic rhinitis ones. A wearable measurement system was developed to record multiple physiological signals to record the onset of non-allergic rhinitis. Vasomotor rhinitis symptoms are generally stimulated by a sudden change in ambient temperature. The temperature of two rooms were set at 20 and 30 ℃, respectively. Subjects were asked to stay in the warm room then enter the cold room to trigger the nasal congestion. A non-allergic rhinitis subject show very different response from that of a Normal subject. The measurement system is to quantitatively record the stimulation and evaluate the treatment result.

M.-S. Jhuang, C.-M. Chen, S.-H. Lu, M.-F. Chen, K.-P. Lin, C.-L. Tsai

Association Between SpO2 Signal Characteristics and Sleep Architecture with Insulin Resistance in Patients with Obstructive Sleep Apnea Syndrome

Obstructive Sleep Apnea Syndrome (OSAS) may contribute to the increasing frequency of metabolic disorders. Intermittent hypoxia (IH) is a major characteristic of the syndrome. However, the existing indices of hypoxia in sleep cannot express accurately the effect of the mild desaturations. In this study, a total of 51 patients without other comorbidities were examined by polysomnography (PSG). Hypoxia parameters were analyzed, in the intervals with low values of SpO2 signal. More specifically, the thresholds were set at 94 and 92% and the average value (M) of the SpO2 signal, in areas below thresholds, were calculated. Moreover, the desaturations were analyzed, together with their duration within the recording in terms of SpO2 signal parameters. The patients’ blood sample was analyzed for metabolic parameters. In total, 28 individuals were diagnosed with severe OSAS, (Apnea Hypopnea index (AHI) 59.11 ± 26.10/h, averSpO2 91.64 ± 4.50%, minSpO2 78.18 ± 10.26%, t < 90 21.42 ± 28.64 and ODI 35.48 ± 33.79/h). A statistically significant correlation between the average M92 value with insulin levels (r = 0.401, p < 0.03) and homeostasis model assessment (HOMA) (r = 0.431, p < 0.022) was displayed. Likewise, a correlation between the amount of desaturations and fasting glucose (r = 0.400, p < 0.035) was observed. Moreover, a statistically significant correlation between the desaturations’ average value with insulin (r = 0.378, p < 0.047) and CRP (r = 0.400, p < 0.035) levels was also revealed. A strong correlation also emerged from the cumulative desaturations’ duration as recorded by the SpO2 signal with fasting glucose levels (r = 0.964 p < 0.001), glycosylated hemoglobin (r = 0.860, p < 0.000) and HOMA index (r = 0.580, p < 0.001). The results suggest that the Hypoxia factors derived from SpO2 signal analysis, are strongly correlated with the insulin resistance and with fasting glucose levels. The correlations of the proposed hypoxia parameters were found to be stronger than the already known hypoxia indices, deriving from the PSG, however a more extended analysis is necessary in order to consolidate the findings of this study.

E. Perantoni, P. Steiropoulos, D. Filos, N. Maglaveras, K. Nikolaou, I. Chouvarda

Biosignal Analysis Methods

Frontmatter

Preprocessing and Filtration Techniques of BSPM Signals in a Small-Scale Study

Cardiac resynchronization therapy (CRT) is an accepted therapeutic option in heart failure treatment. Before the treatment it is necessary to have sufficient information about electrical and mechanical dyssynchrony at each patient. In a joint project of CTU and University Hospital Motol in Prague, data from several modalities are collected and analyzed. In the paper we focus on data preprocessing from multichannel ECG. In particular, we describe in detail the filtration step, as the original signals were contaminated by various artefacts and noise. The aim is to have the signals as clean as possible and preserving the useful information for the analysis.

M. Hrachovina, L. Lhotská, M. Huptych

Blood Vessel Segmentation from Microcirculation Images

Image segmentation is one of the most important steps in analysis of features in image data. In order to acquire stable image of blood vessel, frame-to-frame matching as a step in preprocessing has be made in solving the problem of motion throughout image acquisition. The purpose of this study is to extract the information of the blood vessel from microcirculation video images for the uses in further processes. This image segmentation process had done by dividing the image into two parts, which are the information of the blood vessel and the background. Through the frame-to-frame image intensity pattern based feature analysis and extraction of blood vessel in microcirculation images had used in obtaining the information needed for segmentation. The results showed solid and consistent with the perspective checking result. These segmented blood vessel images has applied for the further measurement of blood flow velocity.

Bea Lyn M. Virtudazo, Jimmy Hasugian, Wen-Chen Lin, Mei-Fen Chen, Kang-Ping Lin

Active Learning for Semi-automated Sleep Scoring

This paper introduces the semi-automatic process using active learning methods which could improve the current state, where a human specialist has to annotate a multiple hours long polysomnographical record to sleep stages. This work is focused on the utilization of density-weighted methods of active learning, one of them turned out to be well-suited for this type of task. Moreover, we proposed several criteria for the comparison of active learning methods. The method saves more than 80% of expert’s annotation effort.

N. Grimova, M. Macas, V. Gerla

Human Fall Detection from Acceleration Measurements Using a Recurrent Neural Network

In this work, a method for human fall detection is presented based on Recurrent Neural Networks. The ability of these networks to process and encode sequential data, such as acceleration measurements from body-worn sensors, makes them ideal candidates for this task. Furthermore, since such networks can benefit greatly from additional data during training, the use of a data augmentation procedure involving random 3D rotations has been investigated. When evaluated on the publicly available URFD dataset, the proposed method achieved better results compared to other methods.

T. Theodoridis, V. Solachidis, N. Vretos, P. Daras

Optimal Threshold Selection for Acceleration-Based Fall Detection

In this paper we present the results of an experiment with 16 subjects performing activities of daily living and simulated falls. We used a triaxial accelerometer to track the subjects’ movements. From the accelerometer data we calculated five different features that are used for fall detection. Contingency tables were built based on the collected dataset and ROC curves were plotted. Optimal thresholds for every feature and corresponding sensitivities and specificities were calculated based on the ROC curve analysis.

G. Šeketa, J. Vugrin, I. Lacković

Camera Based Real Time Fall Detection Using Pattern Classification

A complete real-time fall detection system is presented consisting of camera data acquisition, image processing, pattern recognition, fall alarming, and web interface. Classifiers are trained using only three input features extracted from each video frame using image processing. Among linear and quadratic Bayes, Parzen classifier and 3-nearest neighbors classifier, the last one performed best on a testing set from sensitivity, specificity and area under receiver operating characteristic point of view. Such frame classification is further used in a simple rule triggering the fall alarm process. The fall alarm tested in a real time scenario with 40 falls performed by four persons detected all the falls while having one false positive case.

M. Macaš, S. Lesoin, A. Périn

Machine Learning and Predictive Models in Medicine

Frontmatter

Epileptic Seizures Classification Based on Long-Term EEG Signal Wavelet Analysis

Epilepsy is a complex neurological disorder recognized by abnormal synchronization of cerebral neurons, named seizures. During the last decades, significant progress has been done in automated detection and prediction of seizures, aiming to develop personalized closed-loop intervention systems. In this paper, a methodology for automated seizure detection based on Discrete Wavelet Transform (DWT) is presented. Twenty-one intracranial ictal recordings acquired from the database of University Hospital of Freiburg are firstly segmented in 2 s epochs. Then, a five-level decomposition is applied in each segment and five features are extracted from the wavelet coefficients. The extracted feature vector is used to train a Support Vector Machines (SVM) classifier. Average sensitivity and specificity reached above 93% and 99% respectively.

K. D. Tzimourta, A. T. Tzallas, N. Giannakeas, L. G. Astrakas, D. G. Tsalikakis, M. G. Tsipouras

Heartrate Variability Comparison Between Electrocardiogram, Photoplethysmogram and Ballistic Pulse Waveforms at Fiducial Points

Heart rate variability analysis (HRVA) gives valuable insight to the cardiovascular system. Electrocardiogram (ECG) based HRVA has been assessment gold standard but eavesdropping of wearable technology requires the comparison of its surrogacy to an accepted standard. In this study, optical and mechanical measures at distal artery waveform are compared to the electrical signal of the heart. The sensor data of the six healthy volunteers are collated and compared at fiducial points in various time, frequency and non-linear domains for HRVA. We have found that during early systole fiducial location on waveforms can be surrogate to ECG standard and mechanical sensor 2nd derivative proved to be the best among them. Also, the comparative technology shows enormous potential for cardiovascular diagnostic.

G. M. W. Janjua, R. Hadia, D. Guldenring, D. D. Finlay, J. A. D. McLaughlin

Wavelet ECG Analysis in Time-Frequency Domain of the QRS-Complex in Individuals with Left Bundle Branch Block

LBBB in heart failure patients is a negative predictor for survival. This pattern is also recorded in individuals without significant structural heart diseases. The LBBB morphology has not been previously analysed using wavelet analysis in time-frequency domain in order to identify markers which distinguish these patients. The purpose of this analysis is to investigate if there are any differences in LBBB morphology between normal individuals with LBBB and patients with heart failure and LBBB. Signal-averaged electrocardiograms were recorded, in orthogonal leads and QRS decomposition in nine (9) time-frequency segments was performed using the ‘cmor’ wavelet transformation. Seventy (70) patients (mean age 66, 47 male) were studied. The mean and maximum energies of the QRS complexes were calculated in each of the 9 time-frequency segment. Wavelet parameters of the QRS complex in all segments were higher for normal individuals without LBBB. In the Z lead, the differences in mean QRS energies are significant between the two groups especially in the high frequency band (150–200 Hz) in all time segment whereas the mean energy in X lead presents significant difference in median frequency band (100–150 Hz) in the first time segment. In conclusion wavelet transformation of the QRS complex could differentiate normal individuals from heart failure patients with LBBB.

Kalliopi Papathoma, Stavros Chatzimiltiadis, Nikolaos Maglaveras, Ioanna Chouvarda, Efstratios Theofilogiannakos, Dimitrios Konstantinou, Vassilios Vassilikos

Adaboost Classifier with Dimensionality Reduction Techniques for Epilepsy Classification from EEG

Epilepsy is a serious neurological disorder affecting the human community and this problem has to be dealt with utmost importance. In this disorder, the activity of the neurons in the human brain becomes abnormal and it is witnessed by recurrent seizures. As it is the second most commonly occurring neurological disorder, next to stroke, it affects the quality of life to a great extent. For the clinical evaluation of the activities of the brain, the most commonly used instrument is Electroencephalography (EEG). For dealing with various disorders like classification of epileptic seizures, assessment of mental fatigueness and coma, sleep disorders and schizophrenia, EEG is widely used. As the recordings of the epileptic EEG signal have a very long duration, a lot of data is generated by it. In this paper, the dimensionality of the data is initially reduced with the help of two dimensionality reduction techniques such as Hilbert Transform and Hessian Local Linear Embedding (HLLE). The dimensionally reduced values are then classified with the help of Adaboost Classifier for epilepsy classification from EEG signals. Results show that when Hilbert Transform is classified with Adaboost, a classification accuracy of 93.92% is obtained. When HLLE is classified with Adaboost, a classification accuracy of 92.83% is obtained.

S. K. Prabhakar, H. Rajaguru

Performance Analysis of Factor Analysis and Isomap with Hybrid ABC-PSO Classifier for Epilepsy Classification

In human beings, one of the prevalent and most disturbing neurological disorder is epilepsy which is observed by recurrent seizures due to the abnormal electrical activities of the brain. Due to epilepsy, various symptoms like loss of consciousness, muscle jerks and spasms, fatigueness and so on occur. As the seizures occur randomly, the patients cannot be aware of it and so the physical risk and injury associated with it is quite high. To diagnose and analyze epilepsy, Electroencephalography (EEG) signals are used as an important tool. The recorded scheme of electrical activity due to the excessive firing of neurons within the brain can be measured with the help of EEG. The entire visual analysis of EEG signals is difficult due to its lengthy range of recordings. Therefore, in this paper Factor Analysis and Isomap are used to reduce the dimensions of the EEG data and then the dimensionally reduced values are classified with the help of Hybrid Artificial Bee Colony—Particle Swarm Optimization (ABC-PSO) algorithm to classify the epilepsy from EEG signals. The results show that an average classification accuracy of 96.90% along with an average time delay of 2.23 s is obtained when Factor Analysis is classified with Hybrid ABC-PSO. When Isomap is used along with Hybrid ABC-PSO, a classification accuracy of 97.87% is obtained along with an average time delay of 2.07 s.

S. K. Prabhakar, H. Rajaguru

Performance Analysis of Breast Cancer Classification with Softmax Discriminant Classifier and Linear Discriminant Analysis

The early detection and classification of cancer is very important in order to save the life of a person. One of the dreadful diseases affecting ladies is breast cancer and it is a major concern in the medical field. The breast cancer arises from the tissues of the breast cells. Similar to other parts of the human body, breast comprises of numerous microscopic cells. In the case of breast cancer, the multiplication of the cells happens rapidly in the breast and spreads to other parts of the human body. In the recent years, various machine learning and soft computing techniques were employed to classify various medical issues including breast cancer. In this paper, the breast cancer was classified with the aid of two techniques such as Softmax Discriminant Classifier (SDC) and Linear Discriminant Analysis (LDA). Results show that an average classification accuracy of 97.75% is obtained when LDA is used and an average classification accuracy of 100% is obtained when SDC is used.

S. K. Prabhakar, H. Rajaguru

Behavioural Informatics and Connected Health Technologies

Frontmatter

Emotion Recognition from Haptic Touch on Android Device Screens

The recognition of the emotional state of an individual at a given time point provides valuable information that can find numerous health-related applications, e.g., interventions treating mental or physiological health problems. However, efficient emotion recognition remains a difficult task, often attempted subjectively, with obtrusive means and/or using specialized hardware. The present work uses haptic touch data acquired from Android smartphones to takes the first step towards the development of an objective, unobtrusive and real-life emotion recognition method that exploits the association between emotion and haptic touch. Focusing on four basic emotions (Excitement, Relaxation, Boredom and Frustration) the proposed method achieves very promising classification accuracy using a mixture of feature extraction and machine learning based classification techniques. A well-sized haptic touch dataset has been collected to support the method development and performance evaluation.

C. Maramis, L. Stefanopoulos, I. Chouvarda, N. Maglaveras

Objective Smoking: Towards Smoking Detection Using Smartwatch Sensors

Smoking is arguably one of the most harmful lifestyle behaviors with well-known relations to dozens of diseases. As a consequence, smoking monitoring and cessation support have been popular targets for the emerging field of Behavioral Informatics. Although smoking monitoring has been attempted in the past (e.g., self-reporting), the revolution of smart wearable devices has provided the necessary tools for developing truly objective and unobtrusive smoking detection methods applicable in real-life settings. This work takes the first step towards the development of such a method using smartwatch inertial sensory data, with the end-goal of integrating the proposed method in a novel just-in-time intervention for smoking cessation. The paper explores the detection of smoking instances and the constituting puffs from 3-axis gyroscope data that are routinely acquired by any Android Wear smartphone. The feasibility of the method is tested on actual real-life data, yielding very promising preliminary results.

C. Maramis, V. Kilintzis, P. Scholl, I. Chouvarda

Towards Value Propositions for Persuasive Health and Wellbeing Applications

Recently, considerable attention has been given to health and wellbeing applications, specifically to persuasive applications. Persuasive applications refer to any interactive computing system designed to transform users’ behaviours and attitudes. One of the major challenges of today’s world is that health and wellbeing applications are not sustainable and scientifically designed. However, value proposition (VP) as a denominator might enhance the efficacy of the persuasive health and wellbeing applications. Research has shown little evidence on the VPs in health and wellbeing applications. This paper proposes key VPs for the persuasive health and wellbeing applications. A literature review was conducted based on relevant articles on the value within the health domain. Hence, narrative synthesis literature review approach had been used. We proposed and evaluated these VPs into our built persuasive health and wellbeing applications. We found that the VPs works well with our applications which might enhance their efficacy in the long run.

M. S. Haque, A. Arman, M. Kangas, T. Jämsä, M. Isomursu

Parkinson’s Disease Patients Classification Based on a Motion Tracking Methodology

This study demonstrates how a computer based methodology for tracking motor abilities of Parkinson’s disease can be utilized for patient classification and assessment of the Parkinson’s disease severity. The Line Test methodology evaluates the impaired voluntary movement and generates a set of features that describe the motion. A total cohort of 6 control subjects and 37 Parkinson’s disease subjects were recruited and assessed for the test. During the test, a vertical line appears on the screen and the device evaluates patient’s performance by producing features that correlate the motion to the last medication dosage, the line-test position, the line-test reaction time and the line-test total error. A common cohort of 24 Parkinson’s disease subjects (patients that carried out the Line Test more than once) was formed to track the features alterations between repetitions in time. Results evaluation was performed in both cohorts based on information visualization methodology, optimized for the multi-objective dataset. The line-test position and the time from the last medication dosage features were proved to present the major relation to patients’ group formation. Additionally, line-test reaction time and the line-test total error features proved significant between patients’ performance in the common cohort. Study limitations are correlated to the size of the cohort and the time frame of the study. In general, the current practice supports further investigation into using Line Test methodology for addressing Parkinson’s disease severity.

Eleftheria Polychronidou, Sofia Segkouli, Elias Kalamaras, Stavros Papadopoulos, Anastasios Drosou, Konstantinos Votis, Sevasti Bostantjopoulou, Zoe Katsarou, Charalambos Papaxanthis, Vassilia Hatzitaki, Panagiotis Moschonas, Dimitrios Tzovaras

Patient Empowerment Through Summarization of Discussion Threads on Treatments in a Patient Self-help Forum

Self-help patient fora are widely used for information acquisition and exchange of experiences, e.g., on the effects of medical treatments for a disease. However, a new patient may have difficulties in getting a fast overview of the information inside a large forum. We propose TinnitusTreatmentMonitor, a prototype tool for the summarization and sentiment characterization of postings on medical treatments. We report on applying TinnitusTreatmentMonitor on the platform TinnitusTalk, a self-help platform for tinnitus patients.

Sourabh Dandage, Johannes Huber, Atin Janki, Uli Niemann, Ruediger Pryss, Manfred Reichert, Steve Harrison, Markku Vessala, Winfried Schlee, Thomas Probst, Myra Spiliopoulou

e-Coaching and Patient Support for Physical Activity Promotion

Frontmatter

A Computer-Assisted System with Kinect Sensors and Wristband Heart Rate Monitors for Group Classes of Exercise-Based Rehabilitation

Exercise-based rehabilitation for chronic conditions such as cardiovascular disease, diabetes, and chronic obstructive pulmonary disease, constitutes a key element in reducing patient symptoms and improving health status and quality of life. However, group exercise in rehabilitation programmes faces several challenges imposed by the diversified needs of their participants. In this direction, we propose a novel computer-assisted system enhanced with sensors such as Kinect cameras and wristband heart rate monitors, aiming to support the trainer in adapting the exercise programme on-the-fly, according to identified requirements. The proposed system design facilitates maximal tailoring of the exercise programme towards the most beneficial and enjoyable execution of exercises for patient groups. This work contributes in the design of the next-generation of computerised systems in exercise-based rehabilitation.

A. Triantafyllidis, D. Filos, R. Buys, J. Claes, V. Cornelissen, E. Kouidi, A. Chatzitofis, D. Zarpalas, P. Daras, I. Chouvarda, N. Maglaveras

A Computerized System for Real-Time Exercise Performance Monitoring and e-Coaching Using Motion Capture Data

A lack of exercise and physical activity is one of the main health-risk behaviors, causing chronic diseases. This paper proposes a computerized system for monitoring exercise performance and e-coaching. The aim of the system is to increase users’ physical activity and fitness levels, improve the effectiveness of exercise-based rehabilitation and training, and subsequently motivate people to become more active. Capable of acquiring and fusing motion capture data from different modalities (Kinect and IMUs), depending on the physical exercise intricacy to be captured and evaluated, the proposed system performs real-time anthropometric measurements and analysis. The main challenges of physical exercise performance monitoring are also addressed. Lying-down physical exercises or exercises with caregiver or trainer support can be captured, monitored and evaluated. Exercise repetition detection and evaluation, and e-coaching of the subjects through comprehensive semantic feedback are performed in real-time.

Anargyros Chatzitofis, Dimitris Zarpalas, Petros Daras

Design of a Fully Automated Service to Generate an Individualized Exercise Rehabilitation Program for Adults with Congenital Heart Disease

Physical activity is key in the prevention of cardiovascular diseases and as such should also be targeted in the long-term care of patients with congenital heart disease [CHD]. Mounting evidence, derived from small proof of concept studies, shows that patients with CHD can safely exercise and are able to increase their exercise capacity. Yet, the implementation of physical activity and exercise programs is challenging in patients with CHD given the need for a highly individualized approach. Therefore, efforts are needed to further tailor the current exercise prescription recommendations in order to obtain a truly personalized and most likely more effective exercise rehabilitation program in terms of adherence and health benefits. Here we propose the design of a fully automated service for the generation of an individualized, patient tailored exercise rehabilitation program for adults with CHD. This computer based information system will consider individual patient data to ensure safety, effectiveness as well as attractiveness in order to automatically propose an exercise prescription. This prescription will consist of the short and mid-term exercise goal for the patient to target. Moreover, the system will propose a structured exercise program in order to gradually work toward achieving the goals. The system will support CHD specialists in prescribing detailed, understandable and effective exercise and physical activity programs according to patient needs and wants, based on quantified information on cardiac status, exercise performance, physical activity level and patient preferences. As such, this work contributes to the development of computer-assisted systems targeting at personalized medicine and ultimately better health outcomes in CHD.

R. Buys, V. A. Cornelissen

Adherence to Physical Activity in Patients with Heart Disease: Types, Settings and Evaluation Instruments

Physical fitness is one of the main therapeutic recommendations for patients with heart disease. Yet, adherence to physical activity regimen in both daily life and rehabilitation programs remains to be low. Elaborating on the analysis of factors associated with physical activity behavior, we noticed that the concept of adherence is complex, and its evaluation depends on a specific behavior and its settings. Evaluation of adherence to exercise in leisure time and exercise in the settings of cardiac rehabilitation program requires an application of different instruments. In this paper, we present a summary of findings from the literature analysis regarding types, settings and evaluation instruments of physical activity adherence in daily life and cardiac rehabilitation settings.

K. Livitckaia, V. Koutkias, N. Maglaveras, E. Kouidi, M. van Gils, I. Chouvarda

Training System Methodology Using ECG Signal

The personalized smart training systems is a new technology that supports the personal activity of professional or amateur sport. It is important to avoid overtraining and reach expected results of physical activity during exercises and competitions. One of the most severe health disorders caused by overtraining is the heart disfunction or even heart failure. The electrocardiogram (ECG) signal conveys several different parameters that allow characterizing health conditions and identifying several pathologies. That is why the ECG signal analysis is so important for training. However, during exercise the registration of the signal could be complicated because various noises might appear due to: electrode contact instability, power line interference, baseline wanderings caused by muscular movements and etc. For the analysis the ECG was registered while person was doing squats. This paper contains the description of few methods for ECG data filtering which is done separately: trend (low frequency noise) removal and high frequency noise reduction. This provide more accurate estimation of the signal parameters and it allows to create the training system methodology for decision making. The main task of this research was to describe a training system methodology which includes ECG parameters estimation and training intensity regulations. All these algorithms could be used in future researches for signal compression and improvement of various personalized training systems.

E. Butkeviciute, L. Bikulciene, K. Poderiene

Backmatter

Weitere Informationen