@misc{Dill2018, type = {Master Thesis}, author = {Dill, Sabrina}, title = {Joint Feature Learning and Classification - Deep Learning for Surgical Phase Detection}, url = {http://nbn-resolving.de/urn:nbn:de:0297-zib-81745}, year = {2018}, abstract = {In this thesis we investigate the task of automatically detecting phases in surgical workflow in endoscopic video data. For this, we employ deep learning approaches that solely rely on frame-wise visual information, instead of using additional signals or handcrafted features. While previous work has mainly focused on tool presence and temporal information for this task, we reason that additional global information about the context of a frame might benefit the phase detection task. We propose novel deep learning architectures: a convolutional neural network (CNN) based model for the tool detection task only, called Clf-Net, as well as a model which performs joint (context) feature learning and tool classification to incorporate information about the context, which we name Context-Clf-Net. For the phase detection task lower-dimensional feature vectors are extracted, which are used as input to recurrent neural networks in order to enforce temporal constraints. We compare the performance of an online model, which only considers previous frames up to the current time step, to that of an offline model that has access to past and future information. Experimental results indicate that the tool detection task benefits strongly from the introduction of context information, as we outperform both Clf-Net results and stateof-the-art methods. Regarding the phase detection task our results do not surpass state-of-the-art methods. Furthermore, no improvement of using features learned by the Context-Clf-Net is observed in the phase detection task for both online and offline versions}, language = {en} } @misc{Joachimsky2019, type = {Master Thesis}, author = {Joachimsky, Robert}, title = {Approaching Spinal Kinematics using a Collision-Aware Articulated Deformable Model}, pages = {84}, year = {2019}, abstract = {Statistical Shape Models (SSMs) allow for a compact representation of shape and shape variation and they are a proven means for model-based 3D anatomy reconstruction from medical image data. In orthopaedics and biomechanics, SSMs are increasingly employed to individualize measurement data or to create individualized anatomical models. The human spine is a versatile and complex articulated structure and thus is an interesting candidate to be modeled using an advanced type of SSMs. For modeling and analysis of articulated structures, so called articulated SSMs (aSSMs) have been developed. However, a missing feature of aSSMs is the consideration of collisions in the course of individual fitting and articulation. The aim of this thesis is to develop an aSSM of two adjacent vertebrae that handles collisions between components correctly. The model will incorporate the two major aspects of variability: Shape of a single vertebra and the relative positioning of neighboring vertebrae. That way it becomes possible to adjust shape and articulation in view of a physically and geometrically plausible individualization. To be able to apply collision-aware aSSMs in simulation and optimisation in future work, the approach is based on a parallelized collision detection method employing Graphics Processing Units (GPUs).}, language = {en} } @misc{Neumann2019, type = {Master Thesis}, author = {Neumann, Mario}, title = {Localization and Classification of Teeth in Cone Beam Computed Tomography using 2D CNNs}, url = {http://nbn-resolving.de/urn:nbn:de:0297-zib-74045}, pages = {77}, year = {2019}, abstract = {In dentistry, software-based medical image analysis and visualization provide effcient and accurate diagnostic and therapy planning capabilities. We present an approach for the automatic recognition of tooth types and positions in digital volume tomography (DVT). By using deep learning techniques in combination with dimension reduction through non-planar reformatting of the jaw anatomy, DVT data can be effciently processed and teeth reliably recognized and classified, even in the presence of imaging artefacts, missing or dislocated teeth. We evaluated our approach, which is based on 2D Convolutional Neural Networks (CNNs), on 118 manually annotated cases of clinical DVT datasets. Our proposed method correctly classifies teeth with an accuracy of 94\% within a limit of 2mm distancr to ground truth landmarks.}, language = {en} } @misc{Dill2018, type = {Master Thesis}, author = {Dill, Sabrina Patricia}, title = {Joint Feature Learning and Classification - Deep Learning for Surgical Phase Detection}, pages = {88}, year = {2018}, language = {en} } @misc{Gidey2019, type = {Master Thesis}, author = {Gidey, Henok Hagos}, title = {Automated Hip Knee Ankle Angle Determination using Convolutional Neural Networks}, url = {http://nbn-resolving.de/urn:nbn:de:0297-zib-71263}, pages = {98}, year = {2019}, abstract = {Advanced osteoarthritis is a leading cause of knee replacement and loss of functionality. Early detection of risk factors plays an important role in the application of preventive measures. One of the risk factors is the leg alignment which influences the speed of knee cartilage degradation. The 'gold standard' measurement of leg alignment is done by determining the Hip Knee Ankle (HKA) angle from full lower limb radiographs. Convolutional Neural Networks (CNNs) have gained popularity recently in computer vision. In this thesis we developed methods using CNNs to determine HKA angles from full lower limb radiographs. We trained the CNNs using data from the Osteoarthritis Initiative (OAI). We evaluated our method's performance by evaluating its agreement to experts measurement and its reliability. Our best performing method shows excellent agreement and reliability levels.}, language = {en} } @misc{Reddy2017, type = {Master Thesis}, author = {Reddy, Gutha Vaishnavi}, title = {Automatic Classification of 3D MRI data using Deep Convolutional Neural Networks}, pages = {60}, year = {2017}, abstract = {The chronic disease of Osteoarthritis of the knee that causes pain and discomfort in the knee is associated with the degradation of the joint between the tibia and the femur. The degeneration of this joint is attributed partially to the damage of the meniscus of the knee which forms an important part of the knee joint. Magnetic Resonance Imaging (MRI) is used to diagnose such a kind of osteoarthritis by identifying the degeneration of the knee meniscus. A computer aided diagnostic system that aims to assist a doctor in decision making regarding such a diagnosis can expedite the very diagnosis. Diagnostic decision making for medical imaging falls into the category of classification for a computer vision task. Very Deep Convolutional Networks have been central to the largest advances in computer vision, in recent years. This work entails application of such convolutional networks for the purpose of recognizing a meniscus tear in MRI images as attempting a step towards developing a computer aided diagnosis system for osteoarthritis. Consequently, state-of-the-art pre-trained image recognition networks namely Alexnet, Inceptionv3, VGG and Resnet and Xception were trained on MRI data of the knee meniscus to see if they work for the task of recognizing a tear. A comparison of their classification performance on MRI data was done. The best performing model was the fine-tuned InceptionV3 network which achieved an accuracy close to 60\% for classifying 600 patients based on presence of a tear or not.}, language = {en} } @misc{Tack2015, type = {Master Thesis}, author = {Tack, Alexander}, title = {Gruppenweise Registrierung zur robusten Bewegungsfeldsch{\"a}tzung in artefaktbehafteten 4D-CT-Bilddaten}, year = {2015}, abstract = {Das Ziel der Strahlentherapie ist, eine m{\"o}glichst hohe Dosis in den Tumor zu applizieren und zeitgleich die Strahlenexposition des Normalgewebes zu minimieren. Insbesondere bei thorakalen und abdominalen Tumoren treten aufgrund der Atmung w{\"a}hrend der Bestrahlung große, komplexe und patientenspezifisch unterschiedliche Bewegungen der Gewebe auf. Um den Einfluss dieser Bewegung auf die i.d.R. statisch geplante Dosisverteilung abzusch{\"a}tzen, k{\"o}nnen unter Verwendung der nicht-linearen Bildregistrierung anhand von 3D-CT-Aufnahmen eines Atmungszyklus - also 4D-CT-Daten - zun{\"a}chst die Bewegungsfelder f{\"u}r die strahlentherapeutisch relevanten Strukturen, beispielsweise f{\"u}r die Lunge, berechnet werden. Diese Informationen bilden die Grundlage f{\"u}r sogenannte 4D-Dosisberechnungs- oder Dosisakkumulationsverfahren. Deren Genauigkeit h{\"a}ngt aber wesentlich von der Genauigkeit der Bewegungsfeldsch{\"a}tzung ab. Klassisch erfolgt die Berechnung der Bewegungsfelder mittels paarweiser Bildregistrierung, womit f{\"u}r die Berechnung des Bewegungsfeldes zwischen zwei Bildern im Allgemeinen eine sehr hohe Genauigkeit erreicht wird. Auch f{\"u}r CT-Bilder, die Bewegungsartefakte, wie beispielsweise doppelte oder unvollst{\"a}ndige Strukturen, enthalten, wird unter Verwendung der paarweisen Bildregistrierung im Kontext der Registrierung eine exakte Abbildung der anatomischen Strukturen zwischen den beiden Bildern erreicht. Dabei erfolgt aber eine physiologisch unplausible Anpassung der Felder an die Artefakte. Bei Verwendung der paarweisen Bildregistrierung m{\"u}ssen weiterhin f{\"u}r einen Atemzyklus die Voxel-Trajektorien aus Bewegungsfeldern zwischen mehreren dreidimensionalen Bildern zusammengesetzt werden. Durch Bewegungsartefakte entsprechen diese Trajektorien dann teilweise keiner nat{\"u}rlichen Bewegung der anatomischen Strukturen. Diese Ungenauigkeit stellt in der klinischen Anwendung ein Problem dar; dies gilt umso mehr, wenn Bewegungsartefakte im Bereich eines Tumors vorliegen. Im Gegensatz zu der paarweisen Registrierung kann mit der gruppenweisen Registrierung das Problem der durch Bewegungsartefakte hervorgerufenen ungenauen Abbildung der physiologischen Gegebenheiten dadurch reduziert werden, dass im Registrierungsprozess Bildinformationen aller Bilder, also in diesem Kontext der CT-Daten zu unterschiedlichen Atemphasen, gleichzeitig genutzt werden. Es kann bereits im Registrierungsprozess eine zeitliche Glattheit der Voxel-Trajektorien gefordert werden. In dieser Arbeit wird eine Methode zur B-Spline-basierten zeitlich regularisierten gruppenweisen Registrierung entwickelt. Die Genauigkeit der entwickelten Methode wird f{\"u}r frei zug{\"a}ngliche klinische Datens{\"a}tze landmarkenbasiert evaluiert. Dabei wird mit dem Target Registration Error (TRE) die durchschnittliche dreidimensionale euklidische Distanz zwischen den korrespondierenden Landmarken nach Transformation der Landmarken bezeichnet. Eine Genauigkeit in der Gr{\"o}ßenordnung von aktuellen paarweisen Registrierungen verdeutlicht die Qualit{\"a}t des vorgestellten Registrierungs-Algorithmus. Anschließend werden die Vorteile der gruppenweisen Registrierung durch Experimente an einem Lungenphantom und an manipulierten, artefaktbehafteten klinischen 4D-CT-Bilddaten demonstriert. Dabei werden unter Verwendung der gruppenweisen Registrierung im Vergleich zu der paarweisen Registrierung glattere Trajektorien berechnet, die der realen Bewegung der anatomischen Strukturen st{\"a}rker entsprechen. F{\"u}r die Patientendaten wird außerdem anhand von automatisch detektierten Landmarken der TRE ausgewertet. Der TRE verschlechterte sich f{\"u}r die paarweise Bildregistrierung unter Vorliegen von Bewegungsartefakten von durchschnittlich 1,30 mm auf 3,94 mm. Auch hier zeigte sich f{\"u}r die gruppenweise Registrierung die Robustheit gegen{\"u}ber Bewegungsartefakten und der TRE verschlechterte sich nur geringf{\"u}gig von 1,45 mm auf 1,71 mm.}, language = {de} } @misc{Sahu2016, type = {Master Thesis}, author = {Sahu, Manish}, title = {Instrument Gesture Recognition and Tracking for Effective Control of Laparoscopic Tracking and Guidance Device}, year = {2016}, language = {en} } @misc{vonHafePerezFerreiradaSilva2014, type = {Master Thesis}, author = {von Hafe P{\´e}rez Ferreira da Silva, Maria Isabel}, title = {Computational Left-Ventricle Reconstruction from MRI Data for Patient-specific Cardiac Simulations}, pages = {105}, year = {2014}, language = {en} } @misc{Jalda2009, type = {Master Thesis}, author = {Jalda, Dworzak}, title = {Reconstruction of the Human Rib Cage from 2D Projection Images using a Statistical Shape Model}, year = {2009}, language = {en} } @misc{Renard2011, type = {Master Thesis}, author = {Renard, Maximilien}, title = {Improvement of Image Segmentation Based on Statistical Shape and Intensity Models}, year = {2011}, language = {en} } @misc{Bindernagel2013, type = {Master Thesis}, author = {Bindernagel, Matthias}, title = {Articulated Statistical Shape Models}, year = {2013}, language = {en} } @misc{Zilske2007, type = {Master Thesis}, author = {Zilske, Michael}, title = {Adaptive remeshing of non-manifold triangulations}, year = {2007}, language = {en} } @misc{Weber2008, type = {Master Thesis}, author = {Weber, Britta}, title = {Merkmalskurven auf triangulierten Oberfl{\"a}chen}, year = {2008}, language = {en} } @misc{Wittmers2011, type = {Master Thesis}, author = {Wittmers, Antonia}, title = {Ein Werkzeug zur Erzeugung konsistenter Netze auf triangulierten Oberfl{\"a}chen}, year = {2011}, language = {en} } @misc{Kahnt2012, type = {Master Thesis}, author = {Kahnt, Max}, title = {Generation of constrained high-quality multi-material tetrahedral meshes}, year = {2012}, language = {en} } @misc{Ehlke2012, type = {Master Thesis}, author = {Ehlke, Moritz}, title = {Simulating X-ray images from deformable shape and intensity models on the GPU}, year = {2012}, language = {en} }