Adriana Leal   (UCOMB)
Diana Mendes   (UCOMB)
Diogo Nunes   (UCOMB)
João Pedro Ramos   (UCOMB)
Pierluigi Reali   (POLIMI)
Alvaro Martinez Romero   (UPV)

Adriana Leal is currently attending the PhD program at the Department of Informatics Engineering at the Faculty of Sciences and Technology of University of Coimbra. By enrolling in this program, she intends to contribute with new approaches for epileptic seizure anticipation using neuro-cardiovascular information fusion and dynamic classification.
About 30% of epileptic patients suffer from medically intractable (refractory) epilepsy. Epileptic seizure prediction can be crucial for these patients, who cannot rely on treatment to improve their quality of life. Electroencephalogram (EEG) recordings have been widely used to understand neural activations occurring during epileptic seizures. However, EEG-based approaches often fail to attain real-world applicability, namely, by (1) using low-quality databases comprising discontinuous, short-term EEG signals; (2) proposing EEG features presenting limitations regarding the identification of changes in pre-seizure (a.k.a. pre-ictal) activity; (3) developing prediction approaches that do not take into account personal and context information, which may be indicative of epochs where the patient is more or less prone to the occurrence of seizures.
Additionally, alterations of the cardiovascular status have also been observed in individuals presenting with seizures and, as consequence, more attention has been given to the analysis of electrocardiogram (ECG) signals, often acquired simultaneously with EEG. Fusion of information from both biosignals can be critical in decreasing the number of false positives and significantly improve prediction.
Based on the aforementioned, her PhD project was designed aiming at increasing seizure prediction performance by focusing on four main aspects: (1) exploitation of an appropriate database, (2) extraction of new features, sensitive to pre-seizure changes, (3) EEG and ECG data fusion and (4) formulation of personalised and adaptive methods for seizure prediction that consider variations on the prediction scenario induced by alterations of the patient’s normal body functions, including the vigilance state, patient sleep quality and medication.
The project will be supported on the European Epilepsy Database, which is the largest epilepsy database known so far, concerning number of patients (275 patients), recordings extension and additional patient information. It contains simultaneous EEG and ECG recordings and metadata, such as changes on medication and the vigilance state during seizures.

LINK supported publications
2017
A. Leal, M.G. Ruano, J. Henriques, P. de Carvalho, C. Teixeira, “On the viability of ECG features for seizure anticipation on long-term data”, in 2017 IEEE 3rd International Forum on Research and Technologies for Society and Industry (RTSI), 2017

2018
A. Leal, A. Bianchi, M.G. Ruano, V. Traver, J. Henriques, P. Carvalho, and César Teixeira, “Multisensor data fusion for epileptic seizure prediction: A review of the state of the art”, in Workshop on Innovation on Information and Communication Technologies (ITACA-WIICT 2018), 2018

D. Nunes, A. Leal, T. Rocha, V. Traver, C. Teixeira, S. Paredes, P. Carvalho, J. Henriques, M. Ruano, “Risk prediction of heart failure decompensation events in multiparametric feature spaces”, in 2018 40th International Conference of the IEEE Engineering in Medicine and Biology Society (EMBC), 2018

Diana Mendes is a PhD student in Information Science and Technology (Department of Informatics Engineering in University of Coimbra). The aim of her work is the improvement of clinical diagnosis by integrating knowledge extracted from clinical datasets.
The assessment of a disease or patient condition is a challenging and a decisive task in the daily activities of a physician. Typically, the diagnosis is performed considering the available patient’s information (signs, symptoms, exams, clinical history), supported by the clinical expertise and specific guidelines that helps the physician’s decision. However, there are some problematic conditions where the decision of the professional is not straightforward, in particular, due to the amount and complexity of the information involved. Moreover, there are specific conditions where there is not a complete knowledge and/or the impact of the evolution in modern therapeutics is not totally understood. Therefore, in these cases, the development of appropriated models to help the physician’s decision is recognized to be an essential and a necessary task.
Given the advances in information and communications technology, large clinical datasets resulting from clinical studies, collected during patient hospitalization or gathered using recent tele-monitoring solutions, are nowadays available. These datasets comprise valuable information, which have the potential to be used in the research and in the development of complementary models valuable to support and to help the clinical decision. In effect, mining of these datasets has the potential for establishing new patient stratification principles and for revealing unknown disease correlations.
Based on this information the goal of this work is to derive a systematic approach based on computational intelligence techniques, capable to improve the clinical decision (mainly knowledge based) by means of the integration of additional information extracted from these existent datasets (data driven approaches). According to World Health Organization, cardiovascular diseases are the major cause of death in the world. Given their relevance, the application of two cardiovascular conditions will be carried out in order to validate the proposed strategy:

– risk assessment for coronary artery disease patients;
– prediction of heart failure decompensation for heart failure patients.

LINK supported publications
2017
D. Mendes, S. Paredes, T. Rocha, P. Carvalho, J. Henriques, J. Morais, “Integration of current clinical knowledge with a data driven approach: an innovative perspective”, International Journal of Information Technology & Decision Making, 2017

D. Mendes, S. Paredes, T. Rocha, P. Carvalho, J. Henriques, J. Morais, “Comparison of interpretable data-driven approach with state of the art classifiers: application to cardiovascular risk assessment”, in 2017 IEEE 3rd International Forum on Research and Technologies for Society and Industry (RTSI), 2017

D. Mendes, S. Paredes,T. Rocha, P. Carvalho, J. Henriques, J. Morais, “An interpretable data-driven approach for rules construction: application to cardiovascular risk assessment”, in 39th Int. Conf. of the IEEE Engineering in Medicine and Biology Society, 2017

Diogo Nunes is currently attending the PhD program at the Department of Informatics Engineering of University of Coimbra. His research interest is to combine information of several physiological signals in order to obtain more accurate prediction systems in the context of cardiovascular diseases.
Cardiovascular diseases are a major public health concern, and a cause of considerable morbidity and mortality. Therefore, the prediction of severe events is of great importance for professionals, since it provides the adequate tools to diagnose.
Most of the methods that have been proposed for bio-signal processing have been developed considering the signals as independent. Therefore, the techniques are applied for each signal without taking into account the possible existence of multiple channels of the same signal, the interaction with other signals or even with other types of information. However, given the possible dependence and, particularly their complementarity, it is of major importance to consider the simultaneous analysis of the several sources of information. In fact, the acquired bio-signals are typically generated by different but complementary physiological mechanisms. If a multi-parametric analysis is considered it is viable to estimate indirectly the value of a signal from the other(s) and it is possible, in general, to improve the accuracy of the diagnosis and to reduce the number of false alarms of clinical decision systems by exploiting the redundancy of the information. Thus, the search for new methodologies capable to combining in a global predictive scheme several bio-signals or other heterogeneous sources of information, has been a topic that has increasingly attracted the interest of the research community.
Two main scientific challenges will be addressed: prediction methods and information fusion schemes, based on computational intelligence methodologies. The main applications will be the prediction of hypertension or decompensation episodes for heart failure patients.

LINK supported publications
2017
D. Nunes, P. Carvalho, J. Henriques, M.G. Ruano, C. Teixeira, “Pattern discovery and similarity assessment for robust Heart Sound Segmentation”, in Engineering in Medicine and Biology Society (EMBC), 2017 39th Annual International Conference of the IEEE, 2017

D. Nunes, P. Carvalho, J. Henriques, T. Rocha, “Multiparametric prediction with application to early detection of cardiovascular events”, in 2017 IEEE 3rd International Forum on Research and Technologies for Society and Industry (RTSI), 2017

2018
D. Nunes, A. Leal, T. Rocha, V. Traver, C. Teixeira, S. Paredes, P. Carvalho, J. Henriques, M. Ruano, “Risk prediction of heart failure decompensation events in multiparametric feature spaces”, in 2018 40th International Conference of the IEEE Engineering in Medicine and Biology Society (EMBC), 2018

João Pedro Ramos is a PhD candidate in the doctoral program of information science and technology at the University of Coimbra, Portugal, with an interest in machine learning.
A formal definition for the machine learning concept could be, as T. Mitchell wrote in ‘Machine Learning’, the following:
“A computer program is said to learn from experience E with respect to some class of tasks T and performance measure P, if its performance at tasks in T, as measured by P, improves with experience E.”
Typically, the learning methods are restricted to the use of experience as a set of data appropriately labelled and used to induce the system. They are commonly called data-driven methods as they only rely on the data driven. The fact is, more often than not, there is more experience available on the given domain that can be applied on the learning method, possibly improving the convergence to a more proper end result.
The main goal of his doctoral research is the development of scalable, incremental and adaptive algorithms proven to positively influence data-driven computational learning processes, pursuing the acceptance of highly critic domains. Critic domains expect learnt models to be accurate, impervious to data mishaps and with an open and easy to interpret reasoning.
The basis of this research is ensemble methods. They are one of the trending state of the art tools, achieving reasonable results with the combination of simpler, expectedly easier to understand, models. What constitutes as the right number of models in an ensemble is problem dependent and, sometimes, unknown. In an attempt to cover all basis, a data scientist is driven to overcrowd an ensemble, leading to too many models responding to a data entry, making the interpretation of ensembles difficult.
This investigation focuses on (1) the integration of external knowledge to bias the learning process of ensembles; (2) facilitating the interpretation of ensemble models by streamlining the final set of models.
The methods developed and implemented are expected to give context awareness to the learning process of the domain and sustain more credible and interpretable information to the user. It is expected an overall improvement on accuracy, faster model induction, data conflict resolution and transparent and decipherable reasoning and output. These are fundamental characteristics for any future tool employed in a highly demanding environment.

Pierluigi Reali is a PhD student in Bioengineering at “Politecnico di Milano” (department of Electronic, Informatics and Bioengineering). He is working on the quantification of emotional responses using indexes extracted from ECG and EEG signals, during the execution of certain cognitive tasks.
The emotions that humans experience when interacting with the environment can change the activity levels of their autonomic and central nervous systems. For example, during a psychological stress condition the activity of the sympathetic nervous system (SNS) becomes dominant over the activity of the parasympathetic one (PNS). Since SNS and PNS are responsible for the regulation of cardio-circulatory system, changes in their activation lead to changes in its state (e.g. variations in heart rate, blood pressure, etc.). The strength of this relationship is demonstrated by the well-established link between several psychological and psychiatric conditions, such as stress and depression, and CVDs (cardiovascular diseases). As a consequence, the ability to predict the manifestation of such mental conditions could contribute to both mental and physiological health of the individual, suggesting the adoption of particular precautions for already affected patients and bringing to more effective prevention for the yet undiscovered ones, therefore reducing treatment costs (which is true especially if analyses are performed on easy-to-acquire signals such as ECG and EEG).
Today wearable devices, such as sensorized t-shirts and heart rate monitors, give us the opportunity to collect data in a natural environment and analyze subjects’ behavior during the whole day. For example, we have the possibility of assessing individuals’ stress level in real work conditions, without the issue of recreating those conditions in a laboratory context. The huge amount of data (big data) collected during several days would definitely help to clarify the link between specific mental conditions and cardiovascular diseases, but also requires the use of technics able to reduce computational cost and complexity. For this reason, the implementation of efficient signal processing algorithms and data mining methods is a fundamental part of Pierluigi’s research program.

LINK supported publications
2017
P. Reali, D. Bettiga, A. Mazzola, L. Lamberti, M. Pillan, S. Cerutti, A.M. Bianchi, “Integrated data analysis for the quantification of emotional responses during video observation”, in 2017 IEEE 3rd International Forum on Research and Technologies for Society and Industry (RTSI), 2017

2018
P. Reali, A. Martinez-Millana, P. de Carvalho, A.M. Bianchi, “Cardiovascular effects of stress and emotions: a brief overview of concepts and assessment methods”, in Workshop on Innovation on Information and Communication Technologies (ITACA-WIICT 2018), 2018

P. Reali, C. Cosentini, P. De Carvalho, V. Traver, A.M. Bianchi, “Towards the development of physiological models for emotions evaluation”, in 2018 40th Annual International Conference of the IEEE Engineering in Medicine and Biology Society (EMBC), pp. 110–113, 2018

Álvaro Martínez Romero is a researcher at Polytechnic University of Valencia and PhD student since December 2016. He started working in Sabien group (R&D groups focused on technologies for the health and wellbeing) in 2004 and since then, he had the opportunity to work in several European projects (from FP6 to H2020) in the field of cardiovascular diseases. In LiNK he has found the perfect ecosystem to finally push his PhD studies, combining the research of cardiovascular diseases and a novel field of research called ‘process mining’.
Process mining is a technique that allows discovering process models from event logs. It is commonly used when nor formal description of the process is available or when there exists a descriptive model and conformance between event observations and such model needs to be checked. Process mining techniques can also be used to improve and optimize existing processes by empowering experts to reduce inaccuracies, detect bottlenecks and allow a better use of resources. The paradigm in which experts are involved in the loop is called Interactive Process Mining. This paradigm allows an expert to use the results from process mining techniques to interactively and iteratively modify the process and assess the changes based on future process observations.
The goal of his PhD work is to demonstrate that the application of these interactive process mining techniques to daily clinical practice can improve the involvement of the medical expert in the extraction of scientific evidence and support its application into real settings. Medical experts provide care to patients based on clinical pathways, which are formal medical procedures to treat a specific condition. However, these pathways are general and don’t take into consideration variables such as patient ethnicity, age, gender or comorbidities. In general daily practice, these pathways require a tailoring to each patient and might lead to a try/error loop when, for example, prescribing medication doses. If there exists a registry of all the actions that have been applied to each patient when following such pathway, it is possible to recreate the process and assess their deviations to the base pathway, and more importantly, detect substantial alternate flows that might lead to an improvement in the pathway.
The PhD work will be focused on pathways related to stroke. Strokes happen when blood flow to the brain stops. Within minutes, brain cells begin to die, so the response time to initiate the treatment is critical. The Hospital General of Valencia will be providing a historic registry of 7 years containing the electronic record of those patients admitted either through emergencies or primary care and diagnosed with stroke. The aim of the analysis is to correlate the response time in the triage of stroke to the severity of the disease, and analyze where in the process that response time can be reduced.