The intensive care unit treats the most critically ill patients in the hospital, and as such the clinical staff in the intensive care unit have to deal with complex, time-sensitive and life-critical situations. Commonly, patients present with multiple organ dysfunctions, require breathing and cardiovascular support, which make diagnosis and treatment even more challenging. As a result, clinical staff are faced with processing large quantities of often confusing information, and have to rely on experience and trial and error. This occurs despite the wealth of cardiovascular metrics that are available to the clinician. Computer models of the cardiovascular system can help enormously in an intensive care setting, as they can take the monitored data, and aggregate it in such a way as to present a clear and understandable picture of the cardiovascular system. With additional help that such systems can provide, diagnosis can be more accurate and arrived at faster, alone with better optimised treatment that can start sooner, all of which results in decreased mortality, length of stay and cost. This thesis presents a model of the cardiovascular system, which mimics a specific patient’s cardiovascular state, based on only metrics that are commonly measured in an intensive care setting. This intentional limitation gives rise to additional complexities and challenges in identifying the model, but do not stand in the way of achieving a model that can represent and track all the important cardiovascular dynamics of a specific patient. One important complication that comes from limiting the data set is need for an estimation for the ventricular time-varying elastance waveform. This waveform is central to the dynamics of the cardiovascular model and is far too invasive to measure in an intensive care setting. This thesis thus goes on to present a method in which the value-normalised ventricular time-varying elastance is estimated from only metrics which are commonly available in an intensive care setting. Both the left and the right ventricular time-varying elastance are estimated with good accuracy, capturing both the shape and timing through the progress of pulmonary embolism and septic shock. For pulmonary embolism, with the algorithm built from septic shock data, a time-varying elastance waveform with median error of 1.26% and 2.52% results for the left and right ventricles respectively. For septic shock, with the algorithm built from pulmonary embolism data, a time-varying elastance waveform with median error of 2.54% and 2.90% results for the left and right ventricles respectively. These results give confidence that the method will generalise to a wider set of cardiovascular dysfunctions. Furthermore, once the ventricular time-varying elastance is known, or estimated to a adequate degree of accuracy, the time-varying elastance can be used in its own right to access valuable information about the state of the cardiovascular system. Due to the centrality and energetic nature of the time-varying elastance waveform, much of the state of the cardiovascular system can be found within the waveform itself. In this manner this thesis presents three important metrics which can help a clinician distinguish between, and track the progress of, the cardiovascular dysfunctions of pulmonary embolism and septic shock, from estimations based of the monitored pressure waveforms. With these three metrics, a clinician can increase or decrease their probabilistic measure of pulmonary embolism and septic shock.

Patients infected with the human immunodeficiency virus (HIV+) are living longer and at heightened risk for developing cardiovascular events (CVEs). Commonly used prediction tools appear to misrepresent their CVE risk to varying degrees and in varying directions. Inclusion of markers of cellular infection, chronic immune activation and/or systemic inflammation into risk models might provide better predictive accuracy. Observational studies assessing the relationship of high-sensitivity C-reactive protein (hs-CRP) to CVE in HIV+ patients have reported inconsistent findings. This review of published studies attempted to determine if the available evidence supports its potential use in new models for stable, treated HIV+ patients. We searched the PubMed database using keywords and combinations of "HIV" AND "cardiovascular risk" AND "CRP". Papers presenting original analyses, associating hs-CRP concentration as an independent variable to hard cardiovascular outcomes (myocardial infarction and cardiovascular death), or to hard CVE as part of a composite endpoint, were included. Five observational studies met inclusion/exclusion criteria for review. Three papers identified an association between elevated hs-CRP and CVE, while two others failed to find any significant association. All reports were heterogeneous in terms of independent variables, controls, and designs. The larger and more rigorous studies, employing higher rates of confounder controls and more objective endpoints in their composites, showed positive associations. Though not conclusive, the preponderance of the evidence at this time supports CRP as a potentially valuable factor to be studied in prospective cardiovascular risk prediction investigations in HIV+ patients.

Developing models of the cardiovascular system that simulates the dynamic behavior of a virtual patient’s condition is fundamental in the medical domain for predictive outcome and hypothesis generation. These models are usually described through Ordinary Differential Equation (ODE). To obtain a patient-specific representative model, it is crucial to have an accurate and rapid estimate of the hemodynamic model parameters. Moreover, when adequate model parameters are found, the resulting time series of state variables can be clinically used for predicting the response to treatments and for non-invasive monitoring. In the Thesis, we address the parameter estimation or inverse modeling, by solving an optimization problem, which aims at minimizing the error between the model output and the target data. In our case, the target data are a set of user-defined state variables, descriptive of a hospitalized specific patient and obtained from time-averaged state variables. The Thesis proposes a comparison of both state-of-the-art and novel methods for the estimation of the underlying model parameters of a cardiovascular simulator Aplysia. All the proposed algorithms are selected and implemented considering the constraints deriving from the interaction with Aplysia. In particular, given the inaccessibility of the ODE, we selected gradient-free methods, which do not need to estimate numerically the derivatives. Furthermore, we aim at having a small number of iterations and objective function calls, since these importantly impact the speed of the estimation procedure, and thus the applicability of the knowledge gained through the parameters at the bedside. Moreover, the Thesis addresses the most common problems encountered in the inverse modeling, among which are the non-convexity of the objective function and the identifiability problem. To assist in resolving the latter issue an identifiability analysis is proposed, after which the unidentifiable parameters are excluded. The selected methods are validated using heart failure data, representative of different pathologies commonly encountered in Intensive Care Unit (ICU) patients. The results show that the gradient-free global algorithms Enhanced Scatter Search and Particle Swarm estimate the parameters accurately at the price of a high number of function evaluations and CPU time. As such, they are not suitable for bedside applications. Besides, the local algorithms are not suitable to find an accurate solution given their dependency on the initial guess. To solve this problem, we propose two methods: the hybrid, and the prior-knowledge algorithms. These methods, by including prior domain knowledge, can find a good solution, escaping the basin of attraction of local minima and producing clinically significant parameters in a few minutes. / Utveckling av modeller av det kardiovaskulära systemet som simulerar det dynamiska beteendet hos en virtuell patients är grundläggande inom det medicinska området för att kunna förutsäga resultat och generera hypoteser. Dessa modeller beskrivs vanligtvis genom Ordinary Differential Equation (ODE). För att erhålla en patientspecifik representativ modell är det viktigt att ha en exakt och snabb uppskattning av de hemodynamiska modellparametrarna. När adekvata modellparametrar har hittats kan de resulterande tidsserierna av tillståndsvariabler dessutom användas kliniskt för att förutsäga svaret på behandlingar och för icke-invasiv övervakning. I avhandlingen behandlar vi parameteruppskattning eller invers modellering genom att lösa ett optimeringsproblem som syftar till att minimera följande felet mellan modellens utdata och måldata. I vårt fall är måldata en uppsättning användardefinierade tillståndsvariabler som beskriver en specifik patient som är inlagd på sjukhus och som erhålls från tidsgenomsnittliga tillståndsvariabler. I avhandlingen föreslås en jämförelse av befintlinga och nya metoder. för uppskattning av de underliggande modellparametrarna i en kardiovaskulär simulator, Aplysia. Alla föreslagna algoritmer är valts och implementerade med hänsyn tagna till de begränsningar som finnis i simulatorn Aplysia. Med tanke på att ODE är otillgänglig har vi valt gradientfria metoder som inte behöver uppskatta derivatorna numeriskt. Dessutom strävar vi efter att ha få interationer och funktionsanrop eftersom dessa påverkar hastigheten på estimeringen och därmed den kliniska användbartheten vid patientbehandling. Avhandlingen behandlas dessutom de vanligaste problemen vid inversmodellering som icke-konvexitet och identifierbarhetsproblem. För att lösa det sistnämnda problemet föreslås en identifierbarhetsanalys varefter de icke-identifierbara parametrarna utesluts. De valda metoderna valideras med hjälp av data om hjärtsvikt som är representativa för olika patologier som ofta förekommer hos Intensive Care Unit (ICU)-patienter. Resultaten visar att de gradientfria globala algoritmerna Enhanced Scatter Search och Particle Swarm uppskattar parametrarna korrekt till priset av ett stort antal funktionsutvärderingar och processortid. De är därför inte lämpliga för tillämpningar vid sängkanten. Dessutom är de lokala algoritmerna inte lämpliga för att hitta en exakt lösning eftersom de är beroende av den ursprungliga gissningen. För att lösa detta problem föreslår vi två metoder: hybridalgoritmer och algoritmer med förhandsinformation. Genom att inkludera tidigare domänkunskap kan dessa metoder hitta en bra lösning som undviker de lokala minimernas attraktionsområde och producerar kliniskt betydelsefulla parametrar på några minuter.

In the medtech industry models of the cardiiovascular systems and simulations are valuable tools for the development of new products ad therapies. The simulator Aplysia has been developed over several decade and is able to replicate a wide range of phenomena involved in the physiology and pathophysiology of breathing and circulation. Aplysia is also able to simulate the hemodynamics phenomena starting from a set of patient model parameters enhancing the idea of a "digital twin", i.e. a patient-specific representative simulation. Having a good starting estimate of the patient model parameters is a crucial aspect to start the simulation. A first estimate can be given by looking at patient monitoring data but medical expertise is required. The goal of this thesis is to address the parameter estimation task by developing machine learning and deep learning model to give an estimate of the patient model parameter starting from a set of time-varying data that we will refers as state variables. Those state variables are descriptive of a specific patient and for our project we will generate them through Aplysia starting from the simulation presets already available in the framework. Those presets simulates different physiologies, from healthy cases to different cardiovascular diseases. The thesis propose a comparison between a machine learning pipeline and more complex deep learning architecture to simultaneously predicting all the model parameters. This task is referred as Multi Target Regression (MTR) so the performances will be assessed in terms of MTR performance metrics. The results shows that a gradient boosting regressor with a regressor-stacking approach achieve overall good performances, still it shows some lack of performances on some target model parameters. The deep learning architectures did not produced any valuable results because of the amount of our data: to deploy deep architectures such as ResNet or more complex Convolutional Neural Network (CNN) we need more simulations then the one that were done for this thesis work. / Simulatorn Aplysia har under flera decennier utvecklats för forskning och FoU inom området kardiovaskulära systemmodeller och simuleringar och kan idag replikera ett brett spektrum av fenomen involverade i andningens och cirkulationens fysiologi och patofysiologi. Aplysia kan också simulera hemodynamiska fenomen med utgångspunkt från en uppsättning patientmodellparametrar och detta förstärker idén om en digital tvilling", det vill säga en patientspecifik representativ simulering. Att ha en bra startuppskattning av patientmodellens parametrar är en avgörande aspekt för att starta simuleringen. En första uppskattning kan ges genom att titta på patientövervakningsdata men medicinsk expertis krävs för tolkningen av sådana data. Målet med denna mastersuppsats är att addressera parameteruppskattningsuppgiften genom att utveckla maskininlärnings-och djupinlärningsmodeller för att erhålla en uppskattning av patientmodellparametrar utgående från en uppsättning tidsvarierande data som vi kommer att referera till som tillståndsvariabler. Dessa tillståndsvariabler är beskrivande för en specifik patient och för vårt projekt kommer vi att generera dem med hjälp av Aplysia med utgångspunkt från de modellförinställningar som redan finns tillgängliga i ramverket. Dessa förinställningar simulerar olika fysiologier, från friska fall till olika hjärt-kärlsjukdomar. Uppsatsen presenterar en jämförelse mellan en maskininlärningspipeline och en mer komplex djupinlärningsarkitektur för att samtidigt förutsäga alla modellparametrar. Denna uppgift bygger på MTR så resulterande prestanda kommer att bedömas i termer av MTR prestationsmått. Resultaten visar att en gradientförstärkande regressor med en regressor-stacking-metod uppnår överlag goda resultat, ändå visar den en viss brist på prestanda på vissa målmodellparametrar. Deep learning-arkitekturerna gav inga värdefulla resultat på grund av den begränsade mängden av data vi kunde generera. För att träna djupa arkitekturer som ResNet eller mer komplexa CNN behöver vi fler simuleringar än den som gjordes för detta examensarbete.

Na prática intervencionista, a ultrassonografia intravascular (USIV) é usada para se obter informações quantitativas e qualitativas do acometimento aterosclerótico, de forma complementar à angiografia. Esta tese teve como objetivos explorar a característica tomográfica do exame de USIV, bem como sua dinâmica dentro do ciclo cardíaco. Para isso, desenvolvemos técnicas de processamento de imagens médicas. Primeiramente, investigamos a reconstrução tridimensional da coronária baseando-nos apenas nas imagens de USIV, ou seja, sem a angiografia, como é feita a reconstrução atualmente. Na análise da dinâmica, fizemos um estudo para dispor volumes da coronária em diferentes fases do ciclo cardíaco de forma que estivessem alinhados espacialmente. Como consequência dos tratamentos propostos anteriormente, realizamos estudos sobre a quantificação de propriedades mecânicas dentro das condições oferecidas no intervalo de um ciclo cardíaco. As metodologias propostas foram aplicadas em simulações numéricas desenvolvidas neste trabalho e em exames reais. Obtivemos resultados compatíveis com os objetivos iniciais para reconstrução tridimensional da USIV em simulações numéricas. Na análise da dinâmica, a reconstrução de volumes em diferentes fases do ciclo e o alinhamento espacial possibilitaram a quantificação da variação setorial de volume da luz do vaso durante o ciclo cardíaco / In percutaneous coronary interventions, intravascular ultrasound (IVUS) examination is used to retrieve quantitative and qualitative information about the atherosclerotic plaque progression, complementary to angiography examination. This thesis has as objectives to explore the tomographic characteristic of the IVUS examination, as well as its dynamics within a cardiac cycle. For that purpose, medical image processing techniques were developed. Firstly, we have investigated how to reconstruct the tridimensional coronary based only on IVUS images, that is, without angiography, as it is done nowadays. Regarding dynamic analysis, we have studied models to build volumes of the coronary in distinct phases of the cardiac cycle in a spatial aligned way. Conversantly, as a consequence of the previous image processing methods, we have studied the quantification of mechanical properties of the vessel wall within a cardiac cycle. The methodologies proposed were applied in numeric phantoms developed in this work and also in real IVUS examinations. As result, tridimensional reconstruction was successful in the numeric phantom approach. In dynamics analysis, the reconstruction in distinct cardiac phases and volumes spatial alignment enabled the quantification of lumen volume variation during the cardiac cycle

Na prática intervencionista, a ultrassonografia intravascular (USIV) é usada para se obter informações quantitativas e qualitativas do acometimento aterosclerótico, de forma complementar à angiografia. Esta tese teve como objetivos explorar a característica tomográfica do exame de USIV, bem como sua dinâmica dentro do ciclo cardíaco. Para isso, desenvolvemos técnicas de processamento de imagens médicas. Primeiramente, investigamos a reconstrução tridimensional da coronária baseando-nos apenas nas imagens de USIV, ou seja, sem a angiografia, como é feita a reconstrução atualmente. Na análise da dinâmica, fizemos um estudo para dispor volumes da coronária em diferentes fases do ciclo cardíaco de forma que estivessem alinhados espacialmente. Como consequência dos tratamentos propostos anteriormente, realizamos estudos sobre a quantificação de propriedades mecânicas dentro das condições oferecidas no intervalo de um ciclo cardíaco. As metodologias propostas foram aplicadas em simulações numéricas desenvolvidas neste trabalho e em exames reais. Obtivemos resultados compatíveis com os objetivos iniciais para reconstrução tridimensional da USIV em simulações numéricas. Na análise da dinâmica, a reconstrução de volumes em diferentes fases do ciclo e o alinhamento espacial possibilitaram a quantificação da variação setorial de volume da luz do vaso durante o ciclo cardíaco / In percutaneous coronary interventions, intravascular ultrasound (IVUS) examination is used to retrieve quantitative and qualitative information about the atherosclerotic plaque progression, complementary to angiography examination. This thesis has as objectives to explore the tomographic characteristic of the IVUS examination, as well as its dynamics within a cardiac cycle. For that purpose, medical image processing techniques were developed. Firstly, we have investigated how to reconstruct the tridimensional coronary based only on IVUS images, that is, without angiography, as it is done nowadays. Regarding dynamic analysis, we have studied models to build volumes of the coronary in distinct phases of the cardiac cycle in a spatial aligned way. Conversantly, as a consequence of the previous image processing methods, we have studied the quantification of mechanical properties of the vessel wall within a cardiac cycle. The methodologies proposed were applied in numeric phantoms developed in this work and also in real IVUS examinations. As result, tridimensional reconstruction was successful in the numeric phantom approach. In dynamics analysis, the reconstruction in distinct cardiac phases and volumes spatial alignment enabled the quantification of lumen volume variation during the cardiac cycle