@phdthesis{Martens2020,
  author    = {Martens, Johannes},
  title     = {Development of an In-Silico Model of the Arterial Epicardial Vasculature},
  doi       = {10.25972/OPUS-18247},
  url       = {http://nbn-resolving.de/urn:nbn:de:bvb:20-opus-182478},
  school      = {Universit{\"a}t W{\"u}rzburg},
  year      = {2020},
  abstract  = {In dynamic CE MR perfusion imaging the passage of an intravenously injected CA bolus through tissue is monitored to assess the myocardial pefusion state. To enable this, knowledge of the shape of CA wash-in through upstream epicardial vessels is required, the so-called AIF. For technical reasons this cannot be quantified directly in the supplying vessels and is thus measured in the left ventricle, which introduces the risk of systematic errors in quantification of MBF due to bolus dispersion in coronary vessels. This means occuring CA dispersion must be accounted in the quantification process in order to produce reliable and reproducible results. In order to do this, CFD simulations are performed to analyze and approximate these errors and deepen insights and knowledge gained from previous CFD analyses on both idealized as well as realistic and pathologically altered 3D geometries. In a first step, several different procedures and approaches are undertaken in order to accelerate the performed workflow, however, maintaining a sufficient degree of numerical accuracy. In the end, the implementation of these steps makes the analysis of the cardiovascular 3D model of unprecedented detail including vessels at pre-arteriolar level feasible at all. The findings of the Navier-Stokes simulations are thus validated with regard to different aspects of cardiac blood flow. These include the distribution of VBF into the different myocardial regions, the areals, which can be associated to the large coronary arteries as well as the fragmentation of VBF into vessels of different diameters. The subsequently performed CA transport simulations yield results on the one hand confirming previous studies. On the other hand, interesting additional knowledge about the behavior of CA dispersion in coronary arteries is obtained both regarding travelled distance as well as vessel diameters. The relative dispersion of the so-called vascular transport function, a characterizing feature of vascular networks, shows a linear decrease with vessel diameter. This results in asymptotically decreased additional dispersion of the CA time curve towards smaller and more distal vessels. Nonetheless, perfusion quantification errors are subject to strong regional variability and reach an average value of \$(-28\pm16)\$ \\% at rest across the whole myocardium. Depending on the distance from the inlet and the considered coronary tree, MBF errors up to 62 \\% are observed.},
  subject      = {Computerunterst{\"u}tztes Verfahren},
  language  = {en}
}
@phdthesis{Hahn2010,
  author    = {Hahn, Tim},
  title     = {Integrating neurobiological markers of depression: an fMRI-based pattern classification approach},
  url       = {http://nbn-resolving.de/urn:nbn:de:bvb:20-opus-49962},
  school      = {Universit{\"a}t W{\"u}rzburg},
  year      = {2010},
  abstract  = {While depressive disorders are, to date, diagnosed based on behavioral symptoms and course of illness, the interest in neurobiological markers of psychiatric disorders has grown substantially in recent years. However, current classification approaches are mainly based on data from a single biomarker, making it difficult to predict diseases such as depression which are characterized by a complex pattern of symptoms. Accordingly, none of the previously investigated single biomarkers has shown sufficient predictive power for practical application. In this work, we therefore propose an algorithm which integrates neuroimaging data associated with multiple, symptom-related neural processes relevant in depression to improve classification accuracy. First, we identified the core-symptoms of depression from standard classification systems. Then, we designed and conducted three experimental paradigms probing psychological processes known to be related to these symptoms using functional Magnetic Resonance Imaging. In order to integrate the resulting 12 high-dimensional biomarkers, we developed a multi-source pattern recognition algorithm based on a combination of Gaussian Process Classifiers and decision trees. Applying this approach to a group of 30 healthy controls and 30 depressive in-patients who were on a variety of medications and displayed varying degrees of symptom-severity allowed for high-accuracy single-subject classification. Specifically, integrating biomarkers yielded an accuracy of 83\% while the best of the 12 single biomarkers alone classified a significantly lower number of subjects (72\%) correctly. Thus, integrated biomarker-based classification of a heterogeneous, real-life sample resulted in accuracy comparable to the highest ever achieved in previous single biomarker research. Furthermore, investigation of the final prediction model revealed that neural activation during the processing of neutral facial expressions, large rewards, and safety cues is most relevant for over-all classification. We conclude that combining brain activation related to the core-symptoms of depression using the multi-source pattern classification approach developed in this work substantially increases classification accuracy while providing a sparse relational biomarker-model for future prediction.},
  subject      = {Patientenklassifikation},
  language  = {en}
}