Choose an application

• Reference Manager
• EndNote
• RefWorks (Direct export to RefWorks)

Microneedles are an active field of research in the area of transdermal drug delivery techniques. This thesis aims to build up on the prior fabrication know-how of polymer microneedles. It can be sub-divided into three objectives. Firstly, the thesis project encompasses a further improvement of the existing fabrication process of the SU-8 microneedles. The waste during the fabrication process is furhter reduced: in the original only 20% of the SU8 was actually used, the optimized process this number increased till 95%. Together with this reduction, the overflow during the vacuum oven step is kept under control which leads to a doubling of the SU-8 layer on the wafer and the possibility of quantifying the process parameters. After this optimization a the process parameters are quantified and varied. The second part consists of the development of a generic coating-process for biomolecules (vitamins D2 and B12) onto the existing SU-8 based microneedles without the need for re-formulation of the biomolecule. 2 coatingtechniques for a vitamine B12-coating are explored; spray coating and dipcoating. Dipcoating is the choice of preference because of its fit on a industrial scale. A strong, unform, water suluable coating with a content of 3,14 µg vitamine B12 is achieved with the trial and error technique. Dry electrodes are the last application for the microneedles. First a metal coating is developed and applied with the sputtering tool on the microneedles. Next ECG measurements are done in comparison with the currently used wet electrodes. The results showed an increase of the reciebed amplitude of 1 order of magnitude in the advantage of the microneedles.

Choose an application

• Reference Manager
• EndNote
• RefWorks (Direct export to RefWorks)

Een virtuele autopsie is een techniek binnen forensisch onderzoek om de precieze doodsoorzaak van een kogelschadeslachtoffer te bepalen. In de huidige werkwijze moet een arts om het kogeltraject te bepalen manueel punten van schade aanduiden op 2D sneden. Deze thesis heeft als doel een eerste stap te zetten naar automatisering van dit onderzoek en tracht het schadedetectieproces in 3D CT scans volledig automatisch te doen verlopen. Door intensiteit-gebaseerde vergelijking van een beschadigd beeld met een gesimuleerd pre-mortem beeld, wordt getracht de kogelschade te bepalen. Het pre-mortem beeld stelt een onbeschadigde versie voor van dezelfde patiënt en wordt opgesteld door registratie van een gemiddeld hoofd naar het beschadigde hoofd. De registratie dient anatomische verschillen weg te werken, maar verschillen afkomstig van de kogelschade te negeren. Omdat registratie inherent geen onderscheid kan maken tussen beide verschillen, worden de bekomen vervormingen gecorrigeerd door middel van een principale-componentenanalyse (PCA) model met morfologische voorkennis. Dit model beschrijft vervormingen van het gemiddelde hoofd naar onbeschadigde hoofden. Deze thesis bespreekt verschillende mechanismes voor correctie van vervormingsvelden, die in verschillende mate gebruik maken van het PCA model. Projectie beperkt vervormingen tot deze binnen de PCA vormruimte. Regularisatie gebruikt het model om de bekomen oplossing in de richting van het gemiddelde vervormingsveld te duwen. Het invoeren van gewichten maakt deze regularisatie robuuster door gebieden met lage statistische waarschijnlijkheid te laten domineren door het PCA model, terwijl de vervormingen in andere gebieden voornamelijk bepaald worden door de initiële oplossing van de registratie. Uit foutief gedetecteerde schade in een verschilbeeld met een normaal hoofd blijkt dat de combinatie van gebruikte basis, meer bepaald het gemiddelde hoofd, en de gebruikte registratieparameters leidt tot een imperfecte registratie. Om dezelfde reden worden in het geval van een beschadigd hoofd ook teveel gebieden als beschadigd gemarkeerd. Het verslechteren van het verschilbeeld na toepassing van de voorgestelde correctiemethodes benadrukt het belang van een goed PCA model. Het gebruikte PCA model in deze thesis bleek ontoereikend omdat het teveel variantie verliest aan het modelleren van onbelangrijke of ongewenste structuren, zoals de nek of ruis in de achtergrond. De resultaten tonen echter wel dat de voorgestelde methodes slagen in het corrigeren van het vervormingsveld. Omdat het resultaat van de voorgestelde pipeline zo afhankelijk is van de gebruikte input, namelijk het gemiddeld hoofd, de registratieparameters en het PCA model, worden enkele alternatieven aangeboden waarbij deze input minder zwaar doorweegt.

Choose an application

• Reference Manager
• EndNote
• RefWorks (Direct export to RefWorks)

This thesis aims at modelling the growth plate in endochondral ossification. More precisely, we want to simulate a realistic equilibrium with separate populations of RUNX2 and SOX9, defining a border between hypertrophic cells (in the center) and proliferating cells (at the ends), respectively. Because we study only the first steps of endochondral ossification, no bone growth (the lengthening and broadening), and no vascularisation or ossification centers will be taken in account in the modelling. Also, because of symmetry only 1D models of a halve bone will be elaborated. Firstly some logical models are developed in Matlab, because they are parsimonious and intuitive. A model based on logical equations and a network of Ihh, PTHrP, RUNX2 and SOX9 gives a very rough approximation of the true biological processes, so there is continued with models based on truth tables. A fifth component is added to the truth table, namely BMP2. This model based on truth table with five components is still not satisfactory and thus a model based on partial differential equations (PDEs) is proposed. This PDE model based on reaction diffusion equations with Hill functions and has five components (BMP2, Ihh, PTHrP, RUNX2 and SOX9). This PDE problem is solved with Matlab, with the function pdepe. After a thorough sensitivity analysis, the results are discussed and some comparisons with experimental data are given. The model proves to be adequate, but it is suggested a bigger network is used and the model has to be elaborated more to get more detailed simulations.

Choose an application

• Reference Manager
• EndNote
• RefWorks (Direct export to RefWorks)

Diffusion-weighted magnetic resonance imaging (DWI) is a unique approach to probe the microarchitecture of brain tissue noninvasively. It measures the diffusivity of water molecules, which, in the white matter, is hindered by the the myelin sheath around the axons. Based on this directional information, fiber tracts can be estimated by following the most likely orientations, in a process called tractography.The output of whole brain tractography is a chaos of thousands of fiber trajectories. Clinical applications, however, often demand targeted tracking of specific bundles. Therefore, automated bundle segmentation methods have been developed, which can be categorized into fiber clustering methods and parcellation-based methods. Fiber clustering methods aim to group neighbouring or similar tracks into clusters, which can then be labelled using a white matter atlas. Parcellation-based methods, on the other hand, use regions-of-interest (ROIs) defined in atlas space, e.g., a parcellation of the cortex, to segment and label predefined WM bundles. Tracts that do not meet the requirements, i.e., tracts that do not intersect the ROIs, are excluded from the analysis. This leads to a high specificity but can result in a low sensitivity (a large amount of false negatives). Therefore, this work presents a novel, hybrid approach to fiber bundle segmentation, using spectral embedding and supervised learning. Spectral embedding, a dimensionality reduction technique, allows to represent each track as a vector in the embedding space such that nearby tracks, provided some distance metric, are mapped onto nearby embedding vectors, which simplifies clustering. The embedding vectors calculated on the output of the parcellation-based method are used as ground truth for training.Training data of 20 healthy subjects is labelled with a parcellation-based method, and used to train support vector machine and random forest classifiers. Cross- validation is performed to avoid overfitting. The sensitivity and specificity of the trained classifiers is calculated on labelled data of 5 independent test subjects. This shows an improvement of over 10% when using supervised classifiers. In addition, a clear improvement can be seen in the results on the unlabelled whole brain tractography data which are analysed visually. Moreover, it is demonstrated that by taking the label probabilities into account, the effect of outlier tracks can be reduced.In conclusion, the proposed method outperforms existing unsupervised methods and is more robust to outlier tracks when the label probability is taken into account. Future work includes extending the atlas to smaller, currently undefined fiber bundles.

Choose an application

• Reference Manager
• EndNote
• RefWorks (Direct export to RefWorks)

Prostate cancer is the most common cancer for men, therefore prostate segmentation in MR images has great importance, as it is used for planning treatment therapies. This work focuses on a hybrid prostate segmentation method based on a combination of discriminative learning and a statistical shape model. From a comparison of existing methods, this approach is the most interesting with respect to segmentation time and accuracy. An automated pipeline for localizing and segmenting the prostate in T2 weighted MR images is designed that makes use of a statistical shape model and appearance features in discriminative learning. A discriminative learning approach, based on the face-detection algorithm of Viola-Jones is used to perform the localization task, as well as the search for surface points for the segmentation. In both processes, a big pool of features is used to capture information about appearance. To obtain the final segmentation, the statistical shape model is fitted to the detected surface points. A step-by-step validation study is presented, investigating the influence of the model parameters and the performance of the detection and segmentation steps, separately and combined. Based on the results, a number of interesting observations are obtained, that increase the insight into the approach and give rise to many ideas for future and improving work.