Skip to main content
main-content

Über dieses Buch

In den letzten Jahren hat sich der Workshop "Bildverarbeitung für die Medizin" durch erfolgreiche Veranstaltungen etabliert. Ziel ist auch 2015 wieder die Darstellung aktueller Forschungsergebnisse und die Vertiefung der Gespräche zwischen Wissenschaftlern, Industrie und Anwendern. Die Beiträge dieses Bandes - einige davon in englischer Sprache - umfassen alle Bereiche der medizinischen Bildverarbeitung, insbesondere Bildgebung und -akquisition, Molekulare Bildgebung, Visualisierung und Animation, Bildsegmentierung und - fusion,Sichtbares Licht, Endoskopie, Mikroskoopie, Zeitreihenanalysen, Biomechanische Modellierung, Klinische Anwendung computerunterstützter Systeme, Validierung und Qualitätssicherung, Virtual / Augmented Reality, Bildgestützte Roboter, Chirurgische Simulatoren u.v.m.

Inhaltsverzeichnis

Frontmatter

Complex Motion Modelling in Cancer Imaging

In this talk I will present our recent research efforts and advances in the field of motion modelling of complex organs in oncological applications, as part of the Cancer Research UK / EPSRC Cancer Imaging Centre at Oxford. I will focus on a number of novel non-linear image registration methodologies developed for motion compensation in single- and multi-modality lung imaging, which is a particularly challenging application due to the physiological complexity of the respiration process, such as the interaction between rigid structures, interfacing organs, and large deformations involved. We are currently working on two major challenges in this field: 1. Correcting for the sliding motion of the lungs, by modelling locally discontinuous deformations, and 2. Formulating computational tractable solutions for image alignment between different types of imaging modalities.

Julia Schnabel

Magnetic Particle Imaging

Chancen und Herausforderungen einer neuen Modalität

Magnetic Particle Imaging (MPI) ist ein neues Bildgebungsverfahren, mit dem sich die lokale Konzentration von magnetischen Nanopartikeln quantitativ sowohl mit hoher Empfindlichkeit, als auch mit hervorragender räumlicher Auflösung in Echtzeit darstellen lässt [1]. Diese Vorteile gegenüber etablierten Verfahren, die oft nur einen der Bereiche abdecken können oder nicht quantitativ sind, lassen ein hohes klinisches Potenzial in vielen Anwendungen erwarten. Die Grundidee besteht in der Nutzung der nichtlinearen Magnetisierungskurve der Partikel [2]. Das Verfahren nutzt dazu zwei überlagernde Magnetfelder, zum einen ein statisches Selektionsfeld, zum anderen ein dynamisches Wechselfeld. Werden die Nanopartikel in das Wechselfeld gebracht, erzeugen sie eine nichtlineare Magnetisierung, die mit einer Empfangsspule gemessen werden kann. Aufgrund der Nichtlinearität enthält das gemessene Signal neben der Grundfrequenz des Wechselfelds auch Harmonische, also Schwingungen mit einem Vielfachen der Grundfrequenz. Nach Separation der Harmonischen von dem eingespeisten Grundsignal, kann die Konzentration der Nanopartikel ermittelt werden. Eine örtliche Kodierung wird durch das statische Selektionsfeld erreicht. Als Tracer kommen nanopartikuläre Systeme aus Eisenoxid zum Einsatz. Die Rekonstruktion besteht beim Magnetic Particle Imaging in der Lösung des inversen Problems, bei dem zu den gemessenen induzierten Spannungen, die Konzentrationsverteilung der Nanopartikel bestimmt werden muss. Die Beziehung zwischen beiden Größen wird durch eine entsprechende Systemfunktion beschrieben. Aktuelle Entwicklungen in der Instrumentierung fokussieren insbesondere auf Spulenoptimierungen [4] (Abb. 1) sowie Konzepte für die Ganzkörpertomographie [5].

Thorsten M. Buzug

Data-Driven Spine Detection for Multi-Sequence MRI

Epidemiology studies on vertebra’s shape and appearance require big databases of medical images and image processing methods, that are robust against deformation and noise. This work presents a solution of the first step: the vertebrae detection. We propose a method that automatically detects the central spinal curve with 3D data-driven methods in multi-sequence magnetic resonance images (MRl). Additionally, we use simple edge operations for vertebra border detection that can be used for a statistical evaluation with help of some fast user interaction. Our automatic vertebrae detection algorithm fits a polynomial curve through the spinal canal, that afterwards is shifted towards the vertebra centers. An edge operator gives a first approximation of the vertebra borders, that can be evaluated and corrected by some user interaction within 12 seconds. We show, that our algorithm automatically detects more than 90% of all spines correctly, and present a preliminary analysis of vertebrae sizes.

Daniel Kottke, Gino Gulamhussene, Klaus Tönnies

Automated Breast Volume of Interest Selection by Analysing Breast-Air Boundaries in MRI

The first step in automated breast density estimation is to extract breast volume of interest, namely, the start and end slice numbers from the whole sequence. We evaluated results produced by two radiologists and developed an automatic strategy for the start and end slice detection. The result comparison showed that it is usually more straightforward to find the breast start than the breast end, Where the tissue gradually disappears. In general, the results produced by the algorithm are sufficiently accurate, and our solution will be integrated into a fully automatic breast segmentation pipeline.

Tatyana Ivanovska, Lei Wang, Henry Völzke, Katrin Hegenscheid

Automatische Detektion von Okklusionen zerebraler Arterien in 3D-Magnetresonanzangiographiedaten

Eine schnelle und präzise Detektion verschlossener Hirnarterien ist für die Therapie des ischämischen Schlaganfalls ausschlaggebend. In diesem Beitrag wird eine automatische Methode vorgestellt um okkludierte Blutgefäße in 3D-TOF-MRA-Bildsequenzen aufzufinden. Hierbei werden unter Verwendung verschiedener Schwellwertparameter, auf Grundlage von Vesselnesswerten, alle Endarme des Gefäßskeletts hinsichtlich einer möglichen Okklusion untersucht. Erste Ergebnisse zeigen, dass der vorgestellte Ansatz mit einer Sensitivität von über 85% eine Auffindung verschlossener Arterien ermöglicht und somit eine gute Grundlage für weiterführende Algorithmen darstellt.

Albrecht Kleinfeld, Oskar Maier, Nils Forkert, Heinz Handels

Robust Identification of Contrasted Frames in Fluoroscopic Images

For automatic registration of 3-D models of the left atrium to fluoroscopic images, a reliable classification of images containing contrast agent is necessary. Inspired by previous approaches on contrast agent detection, we propose a learning-based framework which is able to classify contrasted frames more robustly than previous methods, Furthermore, we performed a quantitative evaluation on a clinical data set consisting of 34 angiographies. Our learning-based approach reached a classification rate of 79.5%. The beginning of a contrast injection was detected correctly in 79.4%.

Matthias Hoffmann, Simone Müller, Klaus Kurzidim, Norbert Strobel, Joachim Hornegger

Interaktive und skalierungsinvariante Segmentierung des Rektums/Sigmoid in intraoperativen MRT-Aufnahmen für die gynäkologische Brachytherapie

Gynäkologische Tumore sind die vierthäufigste Art karzinogener Krankheiten. Eine Behandlung besteht i.A. aus Chemotherapie, externer Bestrahlung und interner Strahlentherapie (Brachytherapie). Im Gegensatz zur externen Bestrahlung wird bei der Brachytherapie radioaktives Material direkt in den Tumor oder in seiner unmittelbaren Nähe platziert, Vorher müssen allerdings Tumor und umliegende Organe für eine optimale Strahlendosis segmentiert werden, was – manuell durchgeführt – sehr zeitintensiv ist. In diesem Beitrag stellen wir einen interaktiven, graphbasierten Ansatz zur Segmentierung des Rektums/Sigmoid als ein Risikoorgan (also als Gewebe, das möglichst nicht/wenig bestrahlt werden sollte) der gynäkologischen Brachytherapie vor. Der Ansatz verwendet zur Graphkonstruktion eine benutzerdefinierte Vorlage zur anschließenden interaktiven und skalierungsinvarianten Segmentierung; er wurde anhand von manuellen Segmentierungen von 7 Datensätzen evaluiert, wobei er einen mittleren DSC von 83.85±4,08% und eine mittlere Hausdorff-Distanz von ca. 11 Voxeln erreichte. Im Gegensatz zu einer manuellen Segmentierung, die im Schnitt 5 Minuten dauerte, konnte ein Datensatz mit unserem Ansatz in 2 Minuten segmentiert werden.

Tobias Lüddemann, Jan Egger

Over-Exposure Correction in CT Using Optimization-Based Multiple Cylinder Fitting

Flat-Panel Computed Tomography (CT) has found its commonly used applications in the healthcare field by providing an approach of examining 3D structural information of a human’s body. The popular CT reconstruction algorithms are based on a filtered backprojection (FBP) scheme, which would face challenges when imaging the knee. This because in some views, the X-rays are highly attenuated when traveling through both thigh bones. In the same view, X-rays also travel through soft tissue that absorbs much less energy with respect to bone. When these high intensity X-rays arrive at the detector they cause detector saturation and the generated sinogram suffers from overexposure. Reconstructing an overexposed sinogram results in images with streaking and cupping artifacts, which are unusable for diagnostics. In this paper we propose a method to correct overexposure artifacts using an optimization approach. Parameters describing a specific geometry are determined by thc optimization and then used to extrapolate the overexposed acquisition data.

Alexander Preuhs, Martin Berger, Yan Xia, Andreas Maier, Joachim Hornegger, Rebecca Fahrig

B-Mode-gestützte zeitharmonische Leber-Elastographie zur Diagnose hepatischer Fibrose bei adipösen Patienten

Die Leber-Elastographie ist ein etabliertes bildgebendes Verfahren zur Diagnose hepatischer Fibrose. Limitationen bestehen bei der Untersuchung adipöser Patienten oder Patienten mit Aszites. Daher wurde im Rahmen dieser Arbeit eine neue Methode entwickelt, um anhand zeitharmonischer Ultraschallelastographie (USE) die viskoelastischen Gewebeparameter der Leber in großen und tieferen Messfenstern durchzuführen und eine automatisierte Auswertung zu ermöglichen. Zur Erzeugung von Scherwellen wurde eine Vibrationseinheit in die Patientenliege integriert. Zur Detektion der in die Leber eingekoppelter Scherwellen wurde ein modifizierter klinischer B-Mode-Scanner eingesetzt. Es wurden mehrere Einzelmessungen mit einer Messtiefe von bis zu 14 cm durchgeführt, um die effektive Scherwellengeschwindigkeit sowie den Dispersionsanstieg der Scherwellengeschwindigkeiten zu berechnen. Beide Kenngrößen korrelieren mit der Viskoelastizität des untersuchten Gewebes. Zur Evaluation wurde die USE mit der Magnetresonanzelastographie (MRE) verglichen. Dazu wurden 10 gesunde Freiwilligen mit beiden Verfahren untersucht. Die Resultate zeigen eine sehr gute Übereinstimmung. Die USE wurde nachfolgend in einer klinischen Studie mit 10 weiteren gesunden Freiwilligen und 22 Patienten mit klinisch bewiesener Zirrhose durchgeführt. In Patientenlebern konnte eine signifikante Erhöhung der Scherwellengeschwindigkeiten gegenüber der Kontrollgruppe festgestellt werden. Hiermit wird Medizinern erstmals die nichtinvasive Diagnose der hepatischen Fibrose auf der Grundlage der viskoelastischen Leberveränderungen in adipösen Patienten und bei Vorliegen von Aszites ermöglicht.

Selcan Ipek-Ugay, Heiko Tzschätzsch, Manh Nguyen Trong, Thomas Tolxdorff, Ingolf Sack, Jürgen Braun

Discrete Estimation of Data Completeness for 3D Scan Trajectories with Detector Offset

The sequence of source and detector positions in a CT scan determines reconstructable volume and data completeness. Commonly this is regarded already in the design phase of a scanner. Modern flatpanel scanners, however, allow to acquire a broad range of positions. This enables many possibilities for different scan paths. However, every new path or trajectory implies different data completeness. Analytic solutions are either designed for special trajectories like the Tam-window for helical CT scans or do not incorporate the actual detector size such as Tuy’s condition. In this paper, we describe a method to determine the voxelwise data completeness in percent for discretely sampled trajectories. Doing so, we are able to model any sequence of source and detector positions. Using this method, we are able to confirm known theory such as Tuy’s condition and data completeness of trajectories using detector offset to increase the field-of-view. As we do not require an analytic formulation of the trajectory, the algorithm will also be applicable to any other source-detector-path or set of source-detector-path segments.

Andreas Maier, Patrick Kugler, Günter Lauritsch, Joachim Hornegger

Optimal C-arm Positioning for Aortic Interventions

Due to the continuous integration of innovative interventional imaging modalities into vascular surgery rooms, there is an urgent need for computer assisted interaction and visualization solutions that support the smooth integration of technological solutions within the surgical workflow. In this paper, we introduce a new paradigm for optimalview controlled maneuvering of Angiographic C-arms during thoracic endovascular aortic repair (TEVAR). This allows the semi-automatic pre-computation of well-defined anatomy-related optimal views based on pre-operative 3D image data and automatic interventional positioning of the imaging device relative to the patient’s anatomy through inverse kinematics and CT to patient registration. Together with our clinical partners, we have evaluated the new technique using 5 patient datasets and are able to show promising results.

Salvatore Virga, Verena Dogeanu, Pascal Fallavollita, Reza Ghotbi, Nassir Navab, Stefanie Demirci

Projection and Reconstruction-Based Noise Filtering Methods in Cone Beam CT

A reduction of the radiation dose in computed tomography typically leads to more noise in the acquired projections. Here filtering methods can help to reduce the noise level and preserve the diagnostic value of the low-dose images. In this work, six variants of Gaussian and bilateral filters are applied in both projection and reconstruction domain. Our comparison of noise reduction and image resolution shows that 2D and 3D bilateral filtering in the projection domain can reduce the noise level, but must be applied carefully to avoid streaking artifacts. By smoothing homogeneous regions while preserving sharp edges, the 3D bilateral filter applied in the reconstruction domain yielded the best results in terms of noise reduction and image sharpness.

Benedikt Lorch, Martin Berger, Joachim Hornegger, Andreas Maier

The SIP-NVC-Wizard

User Guidance on Segmentation for Surgical Intervention Planning of Neurovascular Compression

Neurovascular compression syndromes (NVC) result from a compression of the entry/exit zone of cranial nerves by a vessel at the surface of the brainstem. Surgical intervention planning (SIP) for NVC enables a comprehensive spatial understanding of the patient specific anatomy. Specifically, brainstem, vessels and nerves are segmented in NVC SIP. To support segmentation for NVC SIP, we propose a wizardbased user guidance. The SIP-NVC-Wizard guides the user through the patient’s individual anatomy and segmentations, and translates the parameter tuning into adequate medical terminology. We were able to show pathological nerve-vessel-contact in ten NVC cases producing results comparable to conventional NVC SIP, and performed a pilot-study to evaluate usability and functionality of the proposed wizard. With the SIP-NVC-Wizard our subjects were able to perform a NVC SIP without any knowledge in image processing. Anatomical beginners showed a steep learning curve. Time was reduced to 34 minutes compared to 1-2 hours for conventional NVC SIP.

D. Franz, L. Syré, D. Paulus, B. Bischoff, T. Wittenberg, P. Hastreiter

Statistical Analysis of a Qualitative Evaluation on Feature Lines

In this paper, we statistically analyze the results of a qualitative evaluation with 129 participants of the most commonly used feature line techniques on artificial and anatomical structures obtained from patient-specific data. For this analysis, we tested for significant differences to verify the results and the evaluation. We applied the Shapiro-Wilk test for normality, the Friedmann test to validate significant differences, and the Wilcoxon signed-rank test to compare paired samples. The results are used to give recommendations for which kind of surface which feature line technique is most appropriate.

Alexandra Baer, Kai Lawonn, Patrick Saalfeld, Bernhard Preim

Assessment of Electrode Displacement and Deformation with Respect to Pre-Operative Planning in Deep Brain Stimulation

The post-operative validation of deep brain stimulation electrode displacement and deformation is an important task towards improved DBS targeting. In this paper a method is proposed to align models of deep brain stimulation electrodes that are automatically extracted from post-operative CT imaging in a common coordinate system utilizing the planning data as reference. This enables the assessment of electrode displacement and deformation over the whole length of the trajectory with respect to the pre-operative planning. Accordingly, it enables the estimation of plan deviations in the surgical process as well as cross-patient statistics on electrode deformation, e.g. the bending induced by brain-shift.

Andreas Husch, Peter Gemmar, Jörg Lohscheller, Florian Bernard, Frank Hertel

Das 3D User Interface zSpace

Verwendung zur Exploration und Inspektion von Wirbeln der Halswirbelsäule

Diese Arbeit untersucht die Verwendung eines stereoskopischen Monitors sowie die Stift-basierte Eingabe zur Exploration der Halswirbelsäule und Inspektion einzelner Wirbel. Die Exploration medizinischer Strukturen erleichtert das Verstehen und Erlernen anatomischer Zusammenhänge und kann somit Ärzte in der Ausbildung unterstützen. Die Stiftinteraktion basiert auf einer Metapher, welche durch eine Fokusund Kontexttechnik unterstützt wird. Die Eignung des 3D-User Interfaces wird evaluiert sowie quantitativ und qualitativ ausgewertet.

Patrick Saalfeld, Alexandra Baer, Kai Lawonn, Uta Preim, Bernhard Preim

Passive 3D Needle Tip Tracking in Freehand MR-Guided Interventions with Needle Model Visualization

In freehand MR-guided interventions, the monitoring of the current needle position relative to the target is crucial. In this work, a method for fast passive needle tip tracking in 3D is presented. For a true- FISP sequence, it is shown that proper k-space sub-sampling and signal processing allow for an accurate estimation of the needle tip position. A reduction in scan time is achieved by drastically reducing the number of measurements. The calculated needle tip positions are superimposed on a pre-interventional 3D planning volume in form of a needle model to ensure a continuous monitoring of the current needle tip position.

Sebastian Schmitt, Christian Sobotta, Morwan Choli, Heinrich M. Overhoff

Evaluation verschiedener Ansätze zur 4D-4D-Registrierung kardiologischer MR-Bilddaten

4D-MR-Bilddaten des Herzens ermöglichen die Bestimmung diagnostisch relevanter Funktionsparameter. Grundlage für die Berechnung kardiologischer Funktionsparameter sind Segmentierungen des linken bzw. rechten Ventrikels. Die atlasbasierte Segmentierung bietet ein automatisiertes Verfahren, dessen Grundlage nicht-lineare Registrierungsverfahren sind. Dieser Beitrag beschäftigt sich mit der räumlichzeitlichen Registrierung von zwei 4D-Bildsequenzen, auch 4D-4D-Registrierung genannt, durch eine Multichannel-3D-Registrierung mit Trajektorienbeschr änkung. Die Trajektorienbeschränkung bildet korrespondierende Bildpunkte innerhalb einer Sequenz über die Zeit ab und ermöglicht das gleichzeitige Registrieren aller Zeitpunkte zweier 4D-Bildsequenzen durch einen Multichannel-3D-Ansatz. In dieser Arbeit wurde die Multichannel-3D-Registrierung mit weiteren Registrierungsverfahren verglichen und anhand von kardiologischen Cine-MR-Bildsequenzen evaluiert. Es zeigte sich, dass die direkte 3D-Registrierung leicht bessere Ergebnisse erzielte als die Multichannel-3D-Registrierung. Darüber hinaus konnte eine erhöhte Robustheit und Konsistenz durch die Anwendung der Trajektorienbeschränkung festgestellt werden.

Timo Kepp, Jan Ehrhardt, Heinz Handels

Binary Image Inpainting with Interpolation-Enhanced Diffeomorphic Demons Registration

Application to Segmentation Defects of Proximal Femora

There is a wide range of segmentation methods for bone structures in CT images. Many of these methods are declared as automatic, but it is not guaranteed, that the resulting segmentation labels the volume of interest correctly in any case. This work presents a technique, which assists the user with the necessary corrections of the segmentation errors. The procedure must be started manually, but the following steps are fully automatic. First, a similar, correct segmentation is selected from a database, which is used to mask the defects. Then the selected segmentation is registered onto the defect one using the diffeomorphic demons algorithm. Thereby, the region inside the mask is excluded from registration but the displacement field is interpolated. The method has been implemented and tested for segmentations of the proximal femur head, but can easily be transferred to segmentations of other bone regions.

A. Friedberger, O. Museyko, K. Engelke

Handling Non-Corresponding Regions in Image Registration

Image registration is particularly challenging if the images to be aligned contain non-corresponding regions. Using state-of-the-art algorithms typically leads to unwanted and unrealistic deformations in these regions. There are various approaches handling this problem which improve registration results, however each with a focus on specific applications. In this note we describe a general approach which can be applied on different mono-modal registration problems. We show the effects of this approach compared to a standard registration algorithm on the basis of five 3D CT lung image pairs where synthetic tumors have been added. We show that our approach significantly reduces unwanted deformation of a non-corresponding tumor. The average volume decrease is 9% compared to 66% for the standard approach while the overall accuracy based on landmark error is retained.

David Drobny, Heike Carolus, Sven Kabus, Jan Modersitzki

A Memetic Search Scheme for Robust Registration of Diffusion-Weighted MR Images

Effective image-based artifact correction is an essential step in the application of higher order models in diffusion MRI. Most approaches rely on some kind of retrospective registration, which becomes increasingly challenging in the realm of high b-values and low signal-tonoise ratio (SNR), rendering standard correction schemes more and more ineffective. We propose a novel optimization scheme based on memetic search that allows for simultaneous exploitation of different signal intensity relationships between the images, leading to more robust registration results. We demonstrate the increased robustness and efficacy of our method on simulated as well as in-vivo datasets. The median TRE for an affine registration of

b

= 3000 s/mm

2

acquisitions could be reduced from > 5mm for a standard correction scheme to < 1mm using our approach. In-vivo bootstrapping experiments revealed increased precision in all tensor-derived quantities.

Jan Hering, Ivo Wolf, Tawfik Moher Alsady, Hans-Peter Meinzer, Klaus Maier-Hein

A Variational Method for Constructing Unbiased Atlas with Probabilistic Label Maps

We introduce a novel variational method based image registration and reconstruction to construct an average atlas with probabilistic label maps. The average atlas equipped with probabilistic label maps could be used to improve atlas based segmentation. In the experiment we validate the registration accuracy and the unbiasedness of atlas construction using clinical datasets.

Kanglin Chen, Alexander Derksen

MR-Elastographie auf dem Schreibtisch

Untersuchung der Viskoelastizität von Gewebeproben mit einem Niederfeld MR-Tomographen

Bildgebende Verfahren gehören zu den wichtigsten Instrumenten in der medizinischen Diagnostik, während die Palpation ein wichtiges diagnostisches Werkzeug zur qualitativen Erfassung von Elastizitätsveränderungen oberflächennaher Organe ist. Die Elastographie ist die Kombination dieser beiden Techniken. Dabei wird die tastende Hand des Arztes durch niederfrequente Vibrationen und Bildaufnahme der Gewebeauslenkung mittels Ultraschall (US) oder Magnetresonanztomographie (MRT) ersetzt. Mit der Elastographie können viskoelastische Gewebeeigenschaften bestimmt und im komplexen Schermodul G* parametrisiert werden. Die Korrelation der makroskopischen, viskoelastischen Gewebeparameter und der Mikrostruktur ist für eine verbesserte Diagnostik notwendig, sie wurde allerdings bislang nur in wenigen Tierstudien adressiert. Als Lösung bieten sich systematische Untersuchungen von histopathologisch genau charakterisierten Gewebeproben an. Dies scheitert bislang einerseits an der Verfügbarkeit von Magnetresonanz-Elastographie (MRE) tauglichen Geräten, da geeignete Elastogrpahie-Techniken kommerziell noch nicht verfügbar sind. Andererseits ist der Zugang zu MR-Tomographen oftmals limitiert oder mit hohen Kosten verbunden. Im Rahmen dieser Arbeit wurde daher die Machbarkeit der MRE an einem kostengünstigen MRT-Tischgerät mit einem 0.5 Tesla Permanentmagneten untersucht. Für die MRE wurde das Gerät um eine Vibrationseinheit und eine bewegungssensitive Spin-Echo-Aufnahmetechnik erweitert. Die Auswertung erfasster Scherwellenausbreitungsmuster erlaubt die Berechnung von G* und somit die Bestimmung viskoelastischer Gewebeparameter. Die Machbarkeit des Verfahrens wurde mit Agarose- und Ultraschallgel-Proben sowie unterschiedlichen Gewebeproben (Schweineleber, Schweinemuskel und Rinderherz) durchgeführt, wobei sich eine gute Übereinstimmung mit einer Vergleichsuntersuchung an einem Hochfeld Tierscanner gezeigt hat.

Selcan Ipek-Ugay, Toni Drießle, Michael Ledwig, Jing Guo, Thomas Tolxdorff, Ingolf Sack, Jürgen Braun

Portability of TV-Regularized Reconstruction Parameters to Varying Data Sets

In C-arm computed tomography there are certain constraints due to the data acquisition process which can cause limited raw data. The reconstructed image’s quality may significantly decrease depending on these constraints. To compensate for severely under-sampled projection data during reconstruction, special algorithms have to be utilized, more robust to such ill-posed problems. In the past few years it has been shown that reconstruction algorithms based on the theory of compressed sensing are able to handle incomplete data sets quite well. In this paper, the iterative iTV reconstruction method by Ludwig Ritschl et al. is analyzed regarding it’s elimination capabilities of image artifacts caused by incomplete raw data with respect to the settings of it’s various parameters. The evaluation of iTV and the data dependency of iterative reconstruction’s parameters is conducted in two stages. First, projection data with severe angular under-sampling is acquired using an analytical phantom. Proper reconstruction parameters are selected by analyzing the reconstruction results from a set of proposed parameters. In a second step multiple phantom data sets are acquired with limited angle geometry and a small number of projections. The iTV reconstructions of these data sets are compared to short-scan FDK and SART reconstruction results, highlighting the distinct data dependence of the iTV reconstruction parameters.

Mario Amrehn, Andreas Maier, Frank Dennerlein, Joachim Hornegger

Projection-Based Denoising Method for Photon-Counting Energy-Resolving Detectors

In this paper, we present a novel projection-based novel noise reduction method for photon counting energy resolving detectors in Spectral Computed Tomography (CT) imaging. In order to denoise the projection data, a guidance image from all energy channels is computed, which employs a variant of the joint bilateral filter to denoise each energy bin individually. The method is evaluated by a simulation study of cone beam CT data. We achieve a reduction of noise in individual channels by 80% while at the same time preserving edges and structures well in the results, which indicate that the methods are applicable to clinical practice.

Yanye Lu, Michael Manhart, Oliver Taubmann, Tobias Zobel, Qiao Yang, Jang-hwan Choi, Meng Wu, Arnd Doerfler, Rebecca Fahrig, Qiushi Ren, Joachim Hornegger, Andreas Maier

Reference Volume Generation for Subsequent 3D Reconstruction of Histological Sections

Anatomical reference brains are indispensable tools in human brain mapping, enabling the integration of multimodal data or the alignment of a series of adjacent histological brain sections into an anatomically realistic space. This study describes a robust and efficient method for an automatic 3D reconstruction of blockface images taken from postmortem brains during cutting as a prerequisite for high-quality 3D reconstruction of brain sections. The refinement technique used in this registration method is applicable for a broad range of pre-registered histological stacks.

Martin Schober, Philipp Schlömer, Markus Cremer, Hartmut Mohlberg, Anh-Minh Huynh, Nicole Schubert, Mehmet E. Kirlangic, Katrin Amunts

3D Reconstruction of Histological Rat Brain Images

Histology of brain sections provides the opportunity to analyse the brain structure on a microscopic level. Histological sections have to be aligned into a 3D reference volume to obtain a correct spatial localization of cell structures. This paper presents a new approach of a rigid registration method of histological sections to a blockface volume. The method and the results are validated by using synthetic data and experimental histological rat brain images. The results indicate a robust, simple and valid image registration method.

Nicole Schubert, Mehmet E. Kirlangic, Martin Schober, Anh-Minh Huynh, Katrin Amunts, Karl Zilles, Markus Axer

Joint Reconstruction of Multi-Contrast MRI for Multiple Sclerosis Lesion Segmentation

A joint reconstruction framework for multi-contrast MR images is presented and evaluated. The evaluation takes place in function of quality criteria based on reconstruction results and performance in the automatic segmentation of Multiple Sclerosis (MS) lesions. We show that joint reconstruction can effectively recover artificially corrupted images and is robust to noise.

Pedro A Gómez, Jonathan I Sperl, Tim Sprenger, Claudia Metzler-Baddeley, Derek K Jones, Philipp Saemann

Rekonstruktion zerebraler Gefässnetzwerke aus in-vivo μMRA mittels physiologischem Vorwissen zur lokalen Gefässgeometrie

In diesem Beitrag adressieren wir die Rekonstruktion zerebrovaskul ärer Netzwerke mit einem Ansatz, der es erlaubt, Vorwissen über physiologisch plausible Strukturen zu berücksichtigen und gegen- über Bildinformation abzuwägen. Ausgehend von einem überkonnektierten Netzwerk wird in einer globalen Optimierung – unter Berücksichtigung von geometrischer Konstellation, globaler Konnektivität und Bildintensit äten – das plausibelste Netzwerk bestimmt. Ein statistisches Modell zur Bewertung geometrischer Beziehungen zwischen Segmenten und Bifurkationen wird anhand eines hochaufgelösten Netzwerks gelernt, welches aus einem

μ

CT (Mikrocomputertomographie) eines zerebrovaskulären Korrosionspräparats einer Maus gewonnen wird. Die Methode wird experimentell auf in-vivo μMRA (Magnetresonanzmikroangiographie) Datensätze von Mausgehirnen angewandt und Eigenschaften der resultierenden Netzwerke im Vergleich zu Standardverfahren diskutiert.

Markus Rempfler, Matthias Schneider, Giovanna D. Ielacqua, Tim Sprenger, Xianghui Xiao, Stuart R. Stock, Jan Klohs, Gábor Székely, Bjoern Andres, Bjoern H. Menze

Reconstructing a Series of Auto-Radiographic Images in Rat Brains

Quantitative in vitro receptor auto-radiographic studies in brains require the preparation of thin microtome sections. Due to the sectioning process, the spatial coherence is lost and needs to be recovered, if 3D analysis is envisaged. This study describes a new processing pipeline for 3D realignment of auto-radiographs of rat brain sections based on image features. Automatically extracted image features from neighboring sections are matched using their descriptors by rejecting false matches. An intermediate objective is to achieve an intra-subject reconstruction to reduce the manual effort in the next registration step. These steps are followed by a semi-automatic method which aligns already preregistered auto-radiographic stacks into a blockface reference volume to ensure anatomical correctness. The validity of the approach is illustrated by using the mean squared error between the user-defined landmarks as the quality measure.

Anh-Minh Huynh, Mehmet E. Kirlangic, Nicole Schubert, Martin Schober, Katrin Amunts, Karl Zilles, Markus Axer

3D Shape Reconstruction of the Esophagus from Gastroscopic Video

In gastroscopy, video endoscopic imaging is applied for the assessment of the esophagus. Video sequences which provide a narrow two dimensional insight are thereby generated. Three dimensional shape reconstructions from such video sequences offer opportunities for intuitive and enhanced visualization of the esophagus, providing additional contextual and geometrical information. Due to lack of features and the variability of the scene, the shape reconstruction bears a challenge for computer vision. In this contribution, a three dimensional reconstruction from gastroscopic video is presented by first computing a panorama image of the esophagus wall using a novel shape from shading approach followed by a 3D alignment of thereby provided 2D contours of the esophagus wall. The resulting 3D point cloud is then registered contour-wise, leading to a regular triangulation which is then texturized using the panorama image and visualized.

Martin Prinzen, Jonas Trost, Tobias Bergen, Sebastian Nowack, Thomas Wittenberg

Computerunterstützte Planung von Bonebridge Operationen

Zur Unterstützung der präoperativen Planung zur Platzierung eines Hörimplantats am Schädelknochen eines Patienten mit Hörsch ädigung wurde ein Prototyp einer Planungssoftware basierend auf präoperativen DVT-Daten entwickelt. Die Umsetzung erfolgte mittels eines VTK-Widgets, das Manipulationen der Implantatlage sowohl in zweidimensionalen Schnittansichten als auch in einer dreidimensionalen Darstellung des Schädelknochens erlaubt. Zusätzlich wurde auch die Biegung des Implantats berücksichtigt. Dabei lag der Fokus auf einfacher Anwendbarkeit und der Umsetzung geeigneter Manipulationstechniken mittels eines 2D-Eingabegerätes.

M. Scherbinsky, G. J. Lexow, Th. S. Rau, B. Preim, O. Majdani

Real Time Medical Instrument Detection and Tracking in Microsurgery

The detection of surgical instruments in real time is one of the most challenging problems in retinal microsurgery operations. The instrument’s deformable shape, the presence of its shadow, and the illumination variations are the main contributors for such challenge. A new approach for the detection of the tip of the surgical tool is proposed, which can handle the shape deformation, and the presence of the its shadow or the presence of blood vessels. The approach starts by segmenting the tool-like objects using the L*a*b color model. One of these segments is selected as the target tool based on tool’s shaft model. The probabilistic Hough transform was used to get the structural information which can guide us to optimize the best possible candidates’ locations to fit the tool model. The detected tool tip and its slope are propagated between the frames in the images sequence. Experimental results demonstrate the high accuracy of this technique in addition to achieve the real time requirements.

Mohamed Alsheakhali, Mehmet Yigitsoy, Abouzar Eslami, Nassir Navab

Enabling Endovascular Treatment of Type A Dissections

Measurement Scheme for Aortic Surface Lengths

Our goal is to provide a means of enabling minimally invasive therapy within the ascending aorta as a standard clinical procedure in case of aortic dissections. Exact knowledge of the inner and outer surface lengths of the ascending aorta (AA) is essential for producing patientspecific stent-grafts in order to avoid interference with any of the branch vessels connected to the AA. This contribution introduces a genuine approach to quantifying these key parameters. Furthermore, we employ an unique and precise result validation, namely by comparing the accuracy of the inner and outer curvature lengths, determined within a cadaveric CT dataset, with manual measurements performed by a vascular surgeon on the body donor’s excised aorta. Our validated scheme is also being applied under identical circumstances on three patient datasets in pursuance of assessing the variability pertaining to human aortic pathological morphology.

Cosmin Adrian Morariu, Tobias Terheiden, Daniel Sebastian Dohle, Konstantinos Tsagakis, Josef Pauli

Outliers in 3D Point Clouds Applied to Efficient Image-Guided Localization

In this work, the tasks of improving positioning efficiency and minimization of space requirements in image-based navigation are explored. We proved the assumption that it is possible to reduce imagematching time and to increase storage capacities by removing outliers from 3D models used for localization, by applying three outlier removal methods to our datasets and observing the localization associated with the resulting models.

Ekaterina Sirazitdinova, Stephan M. Jonas, Deyvid Kochanov, Jan Lensen, Richard Houben, Hans Slijp, Thomas M. Deserno

Iterative Algorithms to Generate Large Scale Mosaic Images

The process of creating mosaic images from non-linear registration image sequences consists of various complex subtasks. The two most time consuming and hardware resource intensive operations are the image registration process and the solution of an equation system in order to determine image positions in the mosaic. This work presents methods that allow quick calculation of image positions while reducing the necessary hardware resources. A novel graph-based method to determine promising image pairs for the registration process was developed, resulting in a reduction of runtime by 87.5%.

Lorenzo Toso, Stephan Allgeier, Franz Eberle, Susanne Maier, Klaus-Martin Reichert, Bernd Köhler

Variational Registration

A Flexible Open-Source ITK Toolbox for Nonrigid Image Registration

In this article, we present the flexible open-source toolbox “VariationalRegistration” for non-parametric variational image registration, realized as a module in the Insight segmentation and registration toolkit. The toolbox is designed to test, evaluate and systematically compare the effects of different building blocks of variational registration approaches, i.e. the distance/similarity measure, the regularization method and the transformation model. In its current state, the framework includes implementations of different similarity measures and regularization methods, as well as displacement-based and diffeomorphic transformation models. The implementation of further components is possible and encouraged. The implemented algorithms were applied to different registration problems and extensively tested using publicly accessible image data bases. This paper presents a quantitative evaluation for inter-patient registration using 3D brain MR images of the LONI image data base. The results demonstrate that the implemented variational registration scheme is competitive with other state-of-the-art approaches for non-rigid image registration.

Jan Ehrhardt, Alexander Schmidt-Richberg, René Werner, Heinz Handels

Joint Registration and Parameter Estimation of T1 Relaxation Times Using Variable Flip Angles

Accurate and fast estimation of T1 relaxation times is a crucial ingredient for many applications in magnetic resonance imaging [1]. A fast way for T1 estimation is a model-based reconstruction from data obtained with variable flip angles as proposed in [2]. However, this technique requires multiple measurements thus patient movement can degrade the results. In this work we introduce a novel model which combines registration and T1 estimation. A discretization of the new approach is given, including a tailored optimization algorithm. The novel method is compared to conventional reconstruction techniques on 2D software phantom data. With the new method it was possible to improve the relative error in T1 maps from 0.4253 to 0.4049 using the novel algorithm.

Constantin Heck, Martin Benning, Jan Modersitzki

Respiratory Motion Compensation for C-Arm CT Liver Imaging

In C-arm CT 3D liver imaging, breathing leads to motion artifacts due to the relatively long acquisition time. Often, even with breath-holding residual respiratory motion can be observed. These artifacts manifest in blurring and interfere clinical investigations such as liver tissue imaging. For 3D medical image reconstruction a respiratory motion estimation and compensation is required. In this work, the motion was estimated by tracking the motion of the diaphragm and of a vessel bifurcation. The motion signals were integrated into a Thin-Plate-Spline that was used for intra-scan motion compensated reconstruction. This approach was applied to clinical C-arm CT data of the liver and showed improved image quality. Reduced artifacts allow a more precise visual depiction of the liver tissue for liver imaging.

Aline Sindel, Marco Bögel, Andreas Maier, Rebecca Fahrig, Joachim Hornegger, Arnd Dörfler

Detecting Respiratory Artifacts from Video Data

Detecting artifacts in signals is an important problem in a wide number of research areas. In robotic radiotherapy motion prediction is used to overcome latencies in the setup, with robustness effected by the occurrence of artifacts. For motion prediction the detection and especially the definition of artifacts can be challenging. We study the detection of artifacts like, e.g., coughing, sneezing or yawning. Manual detection can be time consuming. To assist manual annotation, we introduce a method based on kernel density estimation to detect intervals of artifacts on video data. We evaluate our method on a small set of test subjects. With 86 intervals of artifacts found by our method we are able to identify all 70 intervals derived from manual detection. Our results indicate a more exact choice of intervals and the identification of subtle artifacts like swallowing, that where missed in the manual detection.

Sven-Thomas Antoni, Robert Plagge, Robert Dürichen, Alexander Schlaefer

Korrektur geometrischer Verzeichnungen zur Kalibrierung von optischen Kohärenztomographiesystemen

Die Optische Kohärenztomographie (OCT) ist ein etabliertes volumetrisches Bildgebungsverfahren, das insbesondere in der Ophthalmologie und Dermatologie angewandt wird. Die vorliegende Arbeit stellt eine neuartige Methode zur Kalibrierung von OCT Systemen vor, die auf Messungen einer selbst angefertigten 3D Referenzstruktur und anschließender landmarkenbasierter Registrierung beruht. Hierdurch sollen geometrische Verzeichnungen korrigiert werden, die insbesondere fehlerhafte Tiefeninformationen liefern. Mit Hilfe unserer Kalibriermethode kann der systematische Fehler um mehr als eine Größenordnung reduziert werden, sodass als Ergebnis quantitative Bildinformationen gewonnen werden können. Dieses Verfahren soll die Rekonstruktion und Interpretation von OCT-Bildern im Hinblick auf medizinische Anwendungen verbessern.

Jenny Stritzel, Jesús Díaz-Díaz, Maik Rahlves, Omid Majdani, Tobias Ortmaier, Eduard Reithmeier, Bernhard Roth

Automatic Single-Cell Segmentation and Tracking of Bacterial Cells in Fluorescence Microscopy Images

Automatic single-cell image analysis allows gaining deeper insights into biological processes. We present an approach for single-cell segmentation and tracking of bacterial cells in time-lapse microscopy image data. For cell segmentation we use linear feature detection and a probability map combined with schemes for cell splitting. For cell tracking we propose an approach based on the maximal overlapping area between cells, which is robust regarding cell rotation and accurately detects cell divisions. Our approach was successfully applied to segment and track cells in time-lapse images of the life cycle of Bacillus subtilis. We also quantitatively evaluated the performance of the segmentation and tracking approaches.

Vaja Liluashvili, Jan-Philip Bergeest, Nathalie Harder, Marika Ziesack, Alper Mutlu, Ilka B. Bischofs, Karl Rohr

Multimodal Image Registration in Digital Pathology Using Cell Nuclei Densities

3D reconstruction and digital double staining offer pathologists many new insights into tissue structure and metabolism. Key to these applications is the precise registration of histological slide images, that is challenging in several ways. One major challenge are differently stained slides, that highlight different parts of the tissue. In this paper we introduce a new registration method to face this multimodality. It abstracts the image information to cell nuclei densities. By minimizing the distance of these densities an affine transformation is determined that restores the lost spatial correspondences. The proposed density based registration is evaluated using consecutive histological slides. It is compared to a Mutual Information based registration and shown to be more accurate and robust.

Nick Weiss, Johannes Lotz, Jan Modersitzki

Räumliche Darstellung und Analyse von Nanopartikelverteilungen in vitalen Alveolarmakrophagen in vitro mit der Dunkelfeldmikroskopie

Die räumliche Wechselwirkung von Nanopartikeln mit vitalen alveolären Makrophagen wurde mit der Dunkelfeldmikroskopie, kombiniert mit einer piezogesteuerten Verschiebung der z-Achse und einer Triggerung der Kamerabilder, untersucht. Damit gelang eine räumliche Rekonstruktion von unfixierten motilen Makrophagen bei gleichzeitiger Darstellung der räumlichen und zeitlichen Verteilung der aufgenommenen Nanopartikel. Die räumliche Darstellung der Nanopartikelverteilungen ermöglicht neue Erkenntnisse zur Biokinetik von Nanopartikeln.

Dominic Swarat, Christian Arens, Martin Wiemann, Hans-Gerd Lipinski

2D Plot Visualization of Aortic Vortex Flow in Cardiac 4D PC-MRI Data

Aortic vortex flow is a strong indicator for various cardiovascular diseases. The correlation of pathologies like bicuspid aortic valves to the occurrence of such flow patterns at specific spatio-temporal positions during the cardiac cycle is of great interest to medical researchers. Dataset analysis is performed manually with common flow visualization techniques such as particle animations. For larger patient studies this is time-consuming and quickly becomes tedious. In this paper, we present a two-dimensional plot visualization of the aorta that facilitates the assessment of occurring vortex behavior at one glance. For this purpose, we explain a mapping of the 4D flow data to circular 2D plots and describe the visualization of the employed λ2-vortex criterion. A grid view allows the simultaneous investigation and comparison of multiple datasets. After a short familiarization with the plots our collaborating cardiologists and radiologists were able distinguish between patient and healthy volunteer datasets with ease.

Benjamin Köhler, Monique Meuschke, Uta Preim, Katharina Fischbach, Matthias Gutberlet, Bernhard Preim

Automatisierung von Vorverarbeitungsschritten für medizinische Bilddaten mit semantischen Technologien

Medizinische Interpretationsverfahren können Ärzte in ihrem täglichen Arbeitsablauf unterstützen, indem Arbeitsschritte im Bereich der Bildvorverarbeitung oder -analyse automatisiert werden. Um dies zu ermöglichen, werden Systeme benötigt, die eigenständig Arbeitsprozesse erstellen und ausführen können. Wir stellen in dieser Arbeit unser Framework anhand des Tumor Progression Mapping (TPM) vor. Es erm öglicht Algorithmen semantisch zu beschreiben und sie automatisch datengetrieben ausführen zu lassen. Wir verwenden dazu Konzepte aus dem Semantic Web: Das Resource Description Framework (RDF) erm öglicht uns Algorithmen mit Semantik anzureichern. Anschließend benutzen wir Linked Data Prinzipien, um eine semantische Architektur zu entwickeln. Wir stellen die Algorithmen als selbstbeschreibende semantische Web Services bereit und führen sie automatisch datengetrieben aus. Wir zeigen anhand dem Tumor Progression Mapping, dass diese deklarative Architektur automatisch verschiedene Arbeitsprozesse erstellen und ausführen kann.

Patrick Philipp, Maria Maleshkova, Michael Götz, Christian Weber, Benedikt Kämpgen, Sascha Zelzer, Klaus Maier-Hein, Miriam Klauß, Achim Rettinger

Towards Standardized Wound Imaging

Self-Calibrating Consumer Hardware Based on Lattice Detection on Color Reference Cards

Photographic documentation in medicine is of increasing importance. Efficient methods are required to properly register and calibrate the images. Usually, a standard reference card with special color pattern is placed in the aperature of the image. Localization and extraction of such cards is a critical step. In this paper, we adopt an iterative lattice detection algorithm developed for outdoor images. Once the lattice is extracted, crossing points of the color fields are used for perspective geometric transform while the color plates guide the color calibration process. Our method is tested on 37 images collected within the German Calciphylaxis Registry. At least, 28 out of the 35 possible grid points have been extracted in all the non-standardizes photographs, with at most two false positive detections. The lowest F-measure was above 80%. Hence, ruler and other calibration devices become obsolete and wound imaging can be performed with low-cost hardware, too.

Abin Jose, Daniel Haak, Stephan M. Jonas, Vincent Brandenburg, Thomas M. Deserno

Implementing a Web-Based Architecture for DICOM Data Capture in Clinical Trials

Medical imaging plays an important role in clinical trials providing qualitative and quantitative findings. Patient’s data in studies is captured in electronic case report forms (eCRFs) instead of paper-based CRFs, which are provided by electronic data capture systems (EDCS). However, EDCS insufficiently support integration of image data into patient’s eCRF. Neither interfacing with picture archiving and communication systems (PACS), nor managing of digital imaging and communications in medicine (DICOM) data in eCRFs is possible. Hence, manual detours for image data in study’s data capture workflow increase errorproneness, latency, and costs. In this work, a completely web-based system architecture is implemented interconnecting EDCS and PACS. Our approach utilizes the open source projects OpenClinica, DCM4CHEE, and Weasis as EDCS, PACS, and DICOM web viewer, respectively. In the optimized workflow, user interaction completely takes place in the eCRF. DICOM data storage and retrieval is performed by middleware components hidden from the user, ensuring data consistency and security by identifier synchronization and de-identification, respectively. This shortens paths for image data capture in the workflow, reduces errors, and saves time and costs. Beside this, valuable data for further research is centrally and anonymously stored in a research PACS.

Daniel Haak, Charles E. Page, Thomas M. Deserno

Semi-automatische Segmentierung von Schädigungszonen in post-interventionellen CT-Daten

Die perkutane Radiofrequenzablation (RFA) ist ein minimalinvasives Verfahren zur thermischen Koagulation von Tumorgewebe und stellt somit eine Alternative zur chirurgischen Entfernung dar. Die Erhitzung wird durch ein elektromagnetisches Wechselfeld erreicht, welches über eine spezielle Nadelanordnung im Gewebe erzeugt wird. Nach der Intervention wird mit Hilfe von CT-Aufnahmen überprüft, inwieweit die Ablation vollständig war, um so das Risiko eines Rezidivs zu minimieren. In diesem Beitrag wurden zwölf RF-Ablationszonen aus post-interventionellen CT-Aufnahmen semiautomatisch segmentiert, um die sehr zeitaufwändige manuelle Inspektion zu unterstützen. Dazu wurde ein interaktiver, graphbasierter Ansatz verwendet, der kugelförmige Objekte bevorzugt. Zur quantitativen und qualitativen Bewertung des Algorithmus wurden manuell segmentierte Schichten von klinischen Experten als Goldstandard verwendet. Zur statistischen Validierung wurde der Dice-Koeffizient herangezogen. Es konnte gezeigt werden, dass der vorgeschlagene Ansatz die Läsionen schneller mit ausreichender Genauigkeit segmentiert und somit für einen Einsatz in der klinischen Routine geeignet zu sein scheint.

Jan Egger, Harald Busse, Michael Moche, Philipp Brandmaier, Daniel Seider, Matthias Gawlitza, Steffen Strocka, Nikita Garnov, Jochen Fuchs, Peter Voigt, Florian Dazinger, Philip Voglreiter, Mark Dokter, Michael Hofmann, Alexander Hann, Bernd Freisleben, Thomas Kahn, Dieter Schmalstieg

Automatic Segmentation of the Cerebral Falx and Adjacent Gyri in 2D Ultrasound Images

We present an automatic segmentation of the cerebral falx and adjacent gyri (perifalcine region) for B-mode 2D ultrasound (US) images. The movement of brain tissue during neurosurgery reduces the accuracy of navigation systems which provide image guidance based on preoperative MRI (preMRI). Thus, the segmentation of the falx and its adjoining gyri in navigated, intraoperative US (iUS) may be used to improve navigation within preMRI scans by providing additional, spatially updated image information of the patient’s brain. The segmentation was tested on 50 2D US images and achieved on average a Dice coefficient of 0.79, a Hausdorff distance of 1.56 mm, and a Jaccard index of 0.64.

Jennifer Nitsch, Jan Klein, Dorothea Miller, Ulrich Sure, Horst K. Hahn

Measurement of the Aortic Diameter in Plain Axial Cardiac Cine MRI

We address the task of aortic diameter measurement in (noncontrast- enhanced) plain axial cardiac cine MRI. To this end, we set up a likelihood maximization problem which allows us to recover globally optimal aorta locations and diameters of the cine sequence efficiently. Our approach provides intuitive means of manual post-correction and requires little user interaction, making large-scale image analysis feasible. Experiments on a data set of 20 cine sequences with 30 time frames showed (at least) pixel-accurate diameter measurements which are also highly stable against re-parameterization.

Marko Rak, Alena-Kathrin Schnurr, Julian Alpers, Klaus-Dietz Tönnies

Detection of Facial Landmarks in 3D Face Scans Using the Discriminative Generalized Hough Transform (DGHT)

This paper presents the Discriminative Generalized Hough Transform (DGHT) as a technique to localize landmarks in 3D face scans. While the DGHT has been successfully used for the detection of landmarks in 2D and 3D images this work extends the framework to be used with triangle meshes for the first time. Instead of edge features and their respective gradient direction, the relative positions and orientations of the mesh faces are utilized to describe the geometric structures which are relevant for the detection of a specific landmark. Implementing a coarse-to-fine strategy at first a decimated version of the mesh is used to locate the global region of the point of interest, followed by more detailed localizations on higher resolution meshes. The utilized shape models are created in an automated, discriminative training process which assigns individual weights to the single model points, aiming at an increased localization rate. The technique has been applied to detect 38 anthropometric facial landmarks on 99 3D face scans. With an average error of 1.9mm, the most accurate detection was performed for the right alare, the average error when considering all landmarks amounts to 5.1 mm.

Gordon Böer, Ferdinand Hahmann, Ines Buhr, Harald Essig, Hauke Schramm

Extraction of the Aortic Dissection Membrane via Spectral Phase Information

Streak/ring/motion artifacts, unequal distribution of the intravenously injected contrast agent and the partial volume effect lead to significant differences in brightness and contrast between CTA datasets or even between slices of the same dataset. These issues affecting the segmentation of fine structures such as the aortic dissection membrane can be efficiently addressed only by applying a measure invariant to luminance and contrast. Towards this end, the analysis of local phase information in the frequency domain using Log-Gabor wavelets achieves sub-mm accuracy when segmenting the dissection membrane by 3 new approaches. In order to avoid under-segmentation, as well as the inclusion of artifacts, this contribution extends phase congruency by proposing a novel scale space strategy, which consists of combining low-frequent, ”secure” structures only with adjoining filtering results of high-frequent nature. Our concept harmonizes with the principles of perceptual organization pertaining to the human visual cortex.

Cosmin Adrian Morariu, Daniel Sebastian Dohle, Konstantinos Tsagakis, Josef Pauli

Fast Adaptive Regularization for Perfusion Parameter Computation

Tuning the Tikhonov Regularization Parameter to the SNR by Regression

Computation of perfusion parameters by deconvolution from contrast-enhanced time-resolved CT or MR perfusion data sets is an illconditioned problem. Thus, adequate regularization and determination of corresponding regularization parameters is required. We present a novel method for Tikhonov regularization for perfusion imaging to locally adapt parameters to the SNR level by using a regression function. In an numerical evaluation our simple approach provided similar or even superior results compared to methods applying computationally more demanding L-curve analysis.

Michael Manhart, Andreas Maier, Joachim Hornegger, Arnd Doerfler

Modellbasierte Simulation der Atembewegung für das Virtual-Reality-Training von Punktionseingriffen

Virtual-Reality-Simulatoren bieten Medizinern eine virtuelle Trainingsumgebung, in der Eingriffe kostengünstig trainiert und geplant werden können, ohne hierbei reale Patienten zu gefährden. Eine Einschränkung der meisten VR-Trainingssimulatoren ist, dass sie von einem statischen Patienten ausgehen, dessen Anatomie im Bereich des simulierten Eingriffs während der Simulation keiner durch die Atmung verursachten Bewegung unterliegt. In diesem Beitrag wird gezeigt, wie Methoden zur Modellierung und Schätzung der Atembewegung aus dem Bereich der Strahlentherapie bewegter Tumoren genutzt werden können, um eine realistische Simulation komplexer, variabler Atembewegungen in VR-Trainingssimulatoren zu erreichen. Die entwickelte Methodik erlaubt eine Visualisierung der Atembewegung in Echtzeit und ermöglicht eine haptische Interaktion mit dem atmenden virtuellen Körper. Dies wird exemplarisch für das Szenario der Leberpunktion gezeigt.

Matthias Wilms, Dirk Fortmeier, André Mastmeyer, Heinz Handels

Rückenschmerz durch Übergewicht?

Biomechanische MKS-Modellierung der Belastungssituation der Lendenwirbelsäule bei unterschiedlichem Körpergewicht

In den letzten Jahrzehnten hat sich Übergewicht und Adipositas zu einem großen globalen gesundheitlichen Problem entwickelt. Während die Auswirkungen auf das kardiovaskuläre System im Fokus vieler Studien stehen, sind die Auswirkungen von Übergewicht und Adipositas auf die Strukturen der Wirbelsäule immer noch nahezu unbekannt. In dieser Studie wurden die Auswirkungen von Normalgewicht und Adipositas mit Hilfe der Mehrkörpersimulation (MKS) auf die Lendenwirbels äule untersucht. Dazu wurden zwei MKS-Modelle, normalgewichtiger Mann und adipöser Mann, erstellt, die sowohl die biomechanischen Eigenschaften der Bandscheiben, der Facettengelenke und der Ligamente berücksichtigen, als auch die anthropometrischen Eigenschaften der zwei Körpergewichtsklassen. Zur Bestimmung der Auswirkungen dieser unterschiedlichen Gewichtsklassen auf die Strukturen der Wirbelsäule, wird die Lendenwirbelsäule jeweils mit der Gewichtskraft eines Normalgewichtigen und eines Adipösen belastet. Dabei werden insbesondere die Belastungsänderungen in den Bandscheiben und in den Facettengelenken untersucht.

Sabine Bauer, Eva Keller, Dietrich Paulus

Markov Random Field-Based Layer Separation for Simulated X-Ray Image Sequences

Motion estimation in X-ray images is a challenging task due to transparently overlapping structures from different depths. We propose to separate an X-ray sequence into a static and a dynamic layer to facilitate motion estimation. The method exploits the idea to use the minimum intensity over time and a spatial smoothness prior for both layers. For numerical optimization, we propose a conditional Markov random field. In experiments on synthetic data, we achieve a root mean squared intensity difference of 36.7±8.4 to the ground truth static layer. In addition, we show qualitative results that demonstrate an improved layer separation compared to state-of-the-art algorithms.

Peter Fischer, Thomas Pohl, Andreas Maier, Joachim Hornegger

Image Registration with Sliding Motion Constraints for 4D CT Motion Correction

A common assumption in medical image registration is that the estimation of a globally continuous deformation field is plausible in reality. However, a sliding behavior of adjacent organ boundaries (e.g. lung and ribcage) cannot be described in a plausible way by a continuous deformation field. In this paper, we address this issue with a novel registration framework that explicitly models sliding of interfaces and can preserve discontinuities in the deformation field along predefined organ boundaries. Incorporated methods involve constrained nonlinear registration and a finite element discretization on unstructured tetrahedral meshes. Evaluation is based on the freely available DIR-Lab datasets.

Alexander Derksen, Stefan Heldmann, Thomas Polzin, Benjamin Berkels

The Cell-Shape-Wizard

User Guidance for Active Contour-Based Cell Segmentation

Cell segmentation on fluorescent micrographs requires preprocessing, cell-background separation and cell-cell separation. The presence of touching or overlapping cells requires more sophisticated segmentation methods – such as Active Contours (AC) – for cell-cell separation, but the usage and parametrization of these methods is often infeasible for users with no image processing expertise. We present the Cell-Shape- Wizard which introduces an abstraction layer between a complex AC approach and the users. It couples tight user guidance with the benefits of interactive cell segmentation of fluorescence micrographs. We have evaluated the wizard in a small user study with four subjects. Results show, that the wizard concept is well applicable to cell segmentation. Segmentation results are compared to manual reference annotations and result in a mean Jaccard index of 0.72. With the Cell-Shape-Wizard life scientist are able to segment their fluorescence micrographs semiautomatically on their own, without being forced to acquire additional knowledge in image processing.

Daniela Franz, H. Huettmayer, Marc Stamminger, Veit Wiesmann, Thomas Wittenberg

Tumorsegmentierung in CD3/CD8-gefärbten Histopathologien

Segmentierung von bestimmten Gewebetypen in Histopathologien ist eine oft untersuchte Fragestellung. Üblicherweise werden dafür Gewebeproben mit Hämatoxylin-Eosin(HE)-Färbung verwendet. CD3/CD8-Färbungen hingegen sind nötig zur Sichtbarmachung von Immunzellen, differenzieren aber nur wenig zwischen unterschiedlichen Gewebearten. Vorteilhaft wäre es, wenn aus nur einem Gewebeschnitt mit einer bestimmten Färbung beide Informationen extrahiert werden könnten. In dieser Arbeit stellen wir ein Segmentierungsverfahren auf CD3/CD8-gefärbten Gewebeproben vor, das effizient zu berechnende und gleichzeitig aussagekräftige Features als Eingabe für einen Clustering- Algorithmus verwendet. In der Evaluation wird ein durchschnittlicher Accuracy-Wert von 94,44% erzielt. Dieser Wert ist vergleichbar mit den Ergebnissen verwandter State of the Art Methoden, die HE-gefärbte Proben einsetzen.

Anqi Wang, Matthias Noll, Stefan Wesarg

Segmentierung von zervikalen Lymphknoten in T1-gewichteten MRT-Aufnahmen

Die Untersuchung von Größe und Aussehen eines Lymphknotens kann ein entscheidender Indikator für die Existenz eines Tumors sein und ist außerdem ein probates Mittel, um Verlaufsanalysen bei einem Patienten durchzuführen, welche wiederum maßgeblichen Einfluss auf die Behandlung haben können. Um die Größe und andere Parameter des Lymphknotens bestimmen zu können, ist zuerst eine Segmentierung vonnöten.Wir präsentieren ein neues Verfahren für die halbautomatische Segmentierung von Lymphknoten auf MR-Datensätzen. Unser Ansatz verwendet eine Wasserscheidentransformation als Grundlage und kombiniert diese mit einem Radialstrahlbasierten Verfahren, um eine möglichst akurate Segmentierung des Lymphknotens zu erhalten. Für die Evaluation wurden 95 Lymphknoten-Segmentierungen aus 17 verschiedenen, kontrastverstärkten T1-gewichteten Patientendatensätzen verwendet. Das durchschnittliche Dice ¨ Ahnlichkeitsmaß lag bei 0.69±0.15 und die mittlere Oberflächendistanz bei 0.65±0.54mm.

Florian Jung, Julia Hilpert, Stefan Wesarg

Dynamic Programming for the Segmentation of Bone Marrow Cells

For the diagnosis of leukemia the morphological analysis of bone marrow is essential. This procedure is time consuming, partially subjective, error-prone and cumbersome. Moreover, repeated examinations may lead to intra- and inter-observer variances. Therefore, an automation of the bone marrow analysis is pursued. The automatic classification of bone marrow cells is highly dependent on the preceding segmentation of the nucleus and plasma parts of the cell. In this contribution we propose a dynamic programming approach for the segmentation of already localized bone marrow cells and evaluate the method with 1000 manually segmented cells. With this approach the segmentation quality for whole cells is 0.93 and 0.85 for the corresponding nucleus parts.

Sebastian Krappe, Christian Münzenmayer, Amrei Evert, Can Fahrettin Koyuncu, Enis Cetin, Torsten Haferlach, Thomas Wittenberg, Christian Held

Colonic Polyp Classification in High-Definition Video Using Complex Wavelet-Packets

In this work, we extend different wavelet-packet based feature extraction methods to use the dual-tree complex wavelet transform. This way we aim at alleviating shortcomings of the different algorithms which stem from the use of the underlying discrete wavelet transform. The derived features are used to classify still-images extracted from HD colonoscopy videos to conduct poly staging. While some techniques cannot benefit from the extension to the dual-tree complex wavelet transform, other benefit in terms of classification accuracy.

M. Häfner, M. Liedlgruber, A. Uhl

Selection of Seeds for Resting-State fMRI-Based Prediction of Individual Brain Maturity

The analysis of resting-state brain connectivity allows unraveling the fundamentals of functional brain organization. Especially changes of network connectivity related to age or diseases promise to serve as early biomarkers. After control of subject movement, we found that, when reaching a critical number of subjects, age prediction is reproducible for all seed selection strategies tested here (functional, anatomical and random based seeds). On the Enhanced Rockland Community Sample, we use support vector regression (SVR) and intense permutation testing for statistical validation.

Norman Scheel, Andrea Essenwanger, Thomas F. Münte, Marcus Heldmann, Ulrike M. Krämer, Amir Madany Mamlouk

Local Surface Estimation from Arbitrary 3D Contour Sets for Aortic Quantification

Possible future endovascular repair of aortic Type A dissections, which always also involve the ascending aorta (AA), implies custom-designed stent-grafts for each patient. Extraction of morphological parameters such as aortic diameters from an accurately reconstructed aortic volume proves essential for accomplishing this goal. Our contribution introduces a novel approach to local surface and normal estimation via implicit models. Given a sparse set of randomly positioned and oriented contours in 3D, we adapt a multi-level partition of unity (MPU) method, rescinding the existing MPU-related restriction of having contours in parallel planes. Radii and diameter measurements are evaluated based on a ground truth aortic volume for 11 patient CTA datasets.

Cosmin Adrian Morariu, Daniel Sebastian Dohle, Tobias Terheiden, Konstantinos Tsagakis, Josef Pauli

Multithreading-Support für die Programmiersprache Julia

Julia ist eine junge, für das wissenschaftliche Rechnen entworfene Programmiersprache. Mit ihrer Matlab verwandten Syntax ist Julia insbesondere für die Forschung im Bereich der medizinischen Bildverarbeitung von Interesse. Trotz einer einfachen Syntax und eines dynamischen Typsystems erreicht Julia Programmlaufzeiten wie äquivalenten C-Programme. Eine Limitierung von Julia ist die fehlende Unterst ützung für das Schreiben von Multithreading-Programmen. In dieser Arbeit stellen wir unser unseren experimentellen Patch vor, der das Ausf ühren von Julia Code in mehreren Threads ermöglicht.

Tobias Knopp

Data-Parallel MRI Brain Segmentation in Clinical Use

Porting FSL-FASTv4 to GPGPUs

Structural MRI brain analysis and segmentation is a crucial part in the daily routine in neurosurgery for intervention planning. Exemplarily, the free software FSL-FAST (FMRIB’s Segmentation Library – FMRIB’s Automated Segmentation Tool) in version 4 is used for segmentation of brain tissue types. To speed up the segmentation procedure by parallel execution, we transferred FSL-FAST to a General Purpose Graphics Processing Unit (GPGPU) using Open Computing Language (OpenCL) [1]. The necessary steps for parallelization resulted in substantially different and less useful results. Therefore, the underlying methods were revised and adapted yielding computational overhead. Nevertheless, we achieved a speed-up factor of 3.59 from CPU to GPGPU execution, as well providing similar useful or even better results.

Joachim Weber, Christian Doenitz, Alexander Brawanski, Christoph Palm

GraphMIC

Medizinische Bildverarbeitung in der Lehre

Die Lehre der medizinischen Bildverarbeitung vermittelt Kenntnisse mit einem breiten Methodenspektrum. Neben den Grundlagen der Verfahren soll ein Gefühl für eine geeignete Ausführungsreihenfolge und ihrer Wirkung auf medizinische Bilddaten entwickelt werden. Die Komplexität der Methoden erfordert vertiefte Programmierkenntnisse, sodass bereits einfache Operationen mit großem Programmieraufwand verbunden sind. Die Software GraphMIC stellt Bildverarbeitungsoperationen in Form interaktiver Knoten zur Verfügung und erlaubt das Arrangieren, Parametrisieren und Ausführen komplexer Verarbeitungssequenzen in einem Graphen. Durch den Fokus auf das Design einer Pipeline, weg von sprach- und frameworkspezifischen Implementierungsdetails, lassen sich grundlegende Prinzipien der Bildverarbeitung anschaulich erlernen. In diesem Beitrag stellen wir die visuelle Programmierung mit GraphMIC der nativen Implementierung äquivalenter Funktionen gegen- über. Die in C++ entwickelte Applikation basiert auf Qt, ITK, OpenCV, VTK und MITK.

Alexander Eduard Szalo, Alexander Zehner, Christoph Palm

A Modular Framework for Post-Processing and Analysis of Fluorescence Microscopy Image Sequences of Subcellular Calcium Dynamics

Calcium (Ca

2+

) signaling is essential for activation of Tlymphocytes and can be understood as fundamental on-switch for the adaptive immune system. The activation is supposed to start by initial spatially and temporally localized Ca

2+

signals. Imaging and analysis of these signals require high spatio-temporal resolution fluorescence microscopy – which, in turn, results in the need for an efficient and reliable post-processing and analysis workflow of the acquired image data. Started with a well established but time-consuming post-processing process, we report on our efforts to automatize and optimize it. The efforts led to a modular post-processing and analysis framework, which is presented. In addition, the influence of instances of the main blocks of the framework (e.g. bleaching correction, deconvolution) on Ca

2+

dynamics analysis measures is evaluated.

Daniel Schetelig, Insa M.A. Wolf, Björn-P. Diercks, Ralf Fliegert, Andreas H. Guse, Alexander Schlaefer, Rene Werner

Automated Whole Slide Analysis of Differently Stained and Co-Registered Tissue Sections

Digital pathology enables applications that are not possible using traditional microscopy and facilitates new ways of handling and presenting whole slide image data, along with quantitative evaluation. Differently stained tissue, highlighting specific biological functions, contains a vast amount of spatial information that must be interpreted by a pathologist. With automated image analysis, some of this information can be quantified and made available for computations such as stain expression analysis. In this contribution we present an automated workflow where quantitative image analysis results of consecutive, differently stained tissue sections are locally fused by co-registration. The results are spatially resolved feature vectors containing features like the densities of positively marked cell types for different stains, which are – in this sense – hyperspectral. Heat maps with many layers (hyperspectral) are generated from this data, revealing relationships between different stains that would not be evident from single stains alone. These hyperspectral data are also a starting point for further investigations; in supporting biomarker discovery in oncology, a systematic search for properties that correlate with clinical data for a patient cohort can be performed in an highly automated way.

Ralf Schönmeyer, Nicolas Brieu, Nadine Schaadt, Friedrich Feuerhake, Günter Schmidt, Gerd Binnig

Band-Pass Filter Design by Segmentation in Frequency Domain for Detection of Epithelial Cells in Endomicroscope Images

Voice hoarseness can have various reasons, one of them is a change of the vocal fold mucus. This change can be examined with micro endoscopes. Cell detection in these images is a difficult task, due to bad image quality, caused by noise and illumination variations. In previous works, it was observed that the repetitive pattern of the cell walls cause an elliptical shape in the Fourier domain [1, 2]. A manual segmentation and back transformation of this shape results in filtered images, where the cell detection is much easier [3]. The goal of this work is to automatically segment the elliptical shape in Fourier domain. Two different approaches are developed to get a suitable band-pass filter: a thresholding and an active contour method. After the band-pass filter is applied, the achieved results are superior to the manual segmentation case.

Bastian Bier, Firas Mualla, Stefan Steidl, Christopher Bohr, Helmut Neumann, Andreas Maier, Joachim Hornegger

Foreground Extraction for Histopathological Whole Slide Imaging

Segmentation of histopathological whole-slide images is a challenging task that requires dedicated approaches. In this paper, the fore- and background segmentation problem is addressed by a combination of basic filters, which is evaluated against the established methods GrabCut and Watershed. It is shown that our computationally efficient, dedicated approach performs better than the technically more advanced methods. The main lesson is that dedicated solutions built on prior knowledge can out-compete advanced algorithms.

Daniel Bug, Friedrich Feuerhake, Dorit Merhof

Sharp as a Tack

Measuring and Comparing Edge Sharpness in Motion-Compensated Medical Image Reconstruction

Organ motion occuring during acquisition of medical images can cause motion blur artifacts, thus posing a major problem for many commonly employed modalities. Therefore, compensating for that motion during image reconstruction has been a focus of research for several years. However, objectively comparing the quality of different motion compensated reconstructions is no easy task. Often, intensity profiles across image edges are utilized to compare their sharpness. Manually positioning such a profile line is highly subjective and prone to bias. Expanding on this notion, we propose a robust, semi-automatic scheme for comparing edge sharpness using an ensemble of profiles. We study the behavior of our approach, which was implemented as an open-source tool, for synthetic data in the presence of noise and artifacts and demonstrate its practical use in respiratory motion-compensated MRI as well as cardiac motion-compensated C-arm CT.

Oliver Taubmann, Jens Wetzl, Günter Lauritsch, Andreas Maier, Joachim Hornegger

Gestenbasierte Interaktionsmethoden für die virtuelle Mikroskopie

Anforderungen und Implementierung in der Pathologie am Beispiel des Leap Motion Sensors

Im Fokus dieses Beitrages liegt die Ermittlung von Anforderungen, die prototypische Implementierung sowie die Verifizierung einer neuartigen Benutzerschnittstelle für die virtuelle Mikroskopie. Zentrale Fragestellungen sind hierbei: Welche Anforderungen werden an die Gestaltung des Arbeitsplatzes eines Pathologen im Nutzungskontext der virtuellen Mikroskopie und insbesondere an die Benutzerschnittstelle des virtuellenMikroskops gestellt? Inwieweit kann eine berührungslose Steuerung diese Vorgaben erfüllen? Zur Beantwortung der Fragen werden Alternativen zur klassischen Bedienung mittels Maus und Tastatur aufgezeigt. Als Methode kommt die nutzerorientierte Gestaltung zum Einsatz, wobei semi-strukturierte Interviews, Personas und Szenarien als Basis für die Ermittlung der Nutzungsanforderungen dienen. Das Ergebnis dieser Arbeit ist eine gestenbasierte Steuerung des virtuellen Mikroskops mittels des Leap Motion Sensors unter Verwendung hierfür entwickelter Gestensets.

Arend Müller, Thorsten Knape, Peter Hufnagl

Spherical Ridgelets for Multi-Diffusion Tensor Refinement

Concept and Evaluation

High angular resolution diffusion imaging (HARDI) improved many neurosurgical areas due to its ability to represent complex intravoxel structures, but is limited for clinical use mainly due to long acquisition times, but also due to noise.

To transcend these limits, our work addresses these problems by combining a state-of-the-art multi diffusion tensor model enhanced with spherical ridgelets. Spherical ridgelets are able to reconstruct a signal based on a limited number of measured directions by utilizing compressed sensing. This concept shows that combining spherical ridgelets with a multi diffusion tensor model can improve the accuracy in case of low signal-to-noise ratios and makes it possible to use less than 15 directional measurements per voxel.

Simon Koppers, Thomas Schultz, Dorit Merhof

Real-Time Resampling of Medical Images Based on Deformed Tetrahedral Structures for Needle Insertion VR-Simulation

To provide real-time visualization of deformed volumetric image data for virtual surgery simulation, a resampling algorithm based on deformed tetrahedral structures has been developed. Deformations of the tetrahedral mesh are computed by a soft-tissue simulation. The major advantage of this approach is the possibility to use the resampled image data in different rendering methods such as ray casting, simulated ultrasound or simulated X-ray imaging. To achieve real-time capability, the algorithm was parallelized on the GPU using Nvidia Cuda. Performance measurements have been done on different mesh resolutions. For a subset of 1688 tetrahedrons, short resamplings times of around 2.3 ms are measured on an Nvidia GTX 680, endorsing the algorithm for real-time application.

Martin Meike, Dirk Fortmeier, Andre Mastmeyer, Heinz Handels

Enhanced Visualization of the Knee Joint Functional Articulation Based on Helical Axis Method

Comprehensive descriptions of the motion in articulating joints open new opportunities in biomedical engineering. The helical axis is a established method that describes flexion-extension at joints, which currently lacks an intuitive visualization. In this paper, we present a comprehensive visualization of knee joint motion based on a direct measurement of the helical axis. The proposed approach incorporates the three-dimensional motion of patient-specific bone segments and the representation of helical axes on the bone, facilitating the observation of the flexion-extension motion at the knee joint.

Ricardo Manuel Millán-Vaquero, Sean Dean Lynch, Benjamin Fleischer, Jan Rzepecki, Karl-Ingo Friese, Christof Hurschler, Franz-Erich Wolter

Panorama Mapping of the Esophagus from Gastroscopic Video

For the examination and clinical assessment of the esophagus, video endoscopy is applied. Video clips and still images are generated along these procedures which are then used for routine documentation. Due to the tight tubular geometry of the esophagus and the constrained field of view of endoscope devices, the provided insight into the esophagus and the relation to contextual information are limited. In this contribution, a shape-from-shading approach for the computation of panorama images of the esophagus wall from gastroscopic video is presented. Furthermore, the content of these panorama images can be mapped back to the original video data which gives the advantages of both panorama-view for improved contextual information and unaltered detail-views for improved examinations.

Martin Prinzen, Martin Raithel, Tobias Bergen, Steffen Mühldorfer, Sebastian Nowack, Dirk Wilhelm, Thomas Wittenberg

Dealing with Intra-Class and Intra-Image Variations in Automatic Celiac Disease Diagnosis

Computer aided celiac disease diagnosis is based on endoscopic images showing the villi structure in regions of the small bowel. Especially unavoidably variable illuminations and varying viewing angles of the individual villi are a source for high intra-class as well as intraimage variations in the image domain. We clarify that common texture descriptors are unable to compensate such a high degree of variance, which is supposed to be a crucial problem in computer aided diagnosis. In this work, a straight-forward split and merge approach is presented which facilitates the final classification task by reducing the intra-image variance and simultaneously enlarging the training set. Using different well known feature extraction techniques as well as two classifiers, it can be shown that the overall classification accuracies can be increased consistently. Additionally, the proposed approach is compared to the related but more complex bag-of-visual-words method.

Michael Gadermayr, Andreas Uhl, Andreas Vécsei

Calibration of Galvanometric Laser Scanners Using Statistical Learning Methods

Galvanometric laser scanners can be used for optical tracking. Model-based calibration of these systems is inaccurate and not adaptable to variations in the system. Therefore, a calibration method based on statistical learning methods is presented which directly incorporates the triangulation problem. We investigate linear regression as well as Artificial Neural Networks. The results are validated using (1) the cross-validated prediction accuracy within the calibration space, and (2) plane reconstruction accuracy. All statistical learning methods outperformed the model-based approach leading to an improvement of up to 74% for the cross-validated 3D root-mean-square error and 70-74% for the plane reconstruction. While the neural network achieved mean errors below 0.5 mm, the linear regression results suggest a good compromise between accuracy and computational load.

Stefan Lüdtke, Benjamin Wagner, Ralf Bruder, Patrick Stüber, Floris Ernst, Achim Schweikard, Tobias Wissel

Überwachtes Lernen zur Prädiktion von Tumorwachstum

In der Bestrahlungsplanung bei Hirntumoren wird typischerweise ein Sicherheitsabstand von 2 − 2, 5 cm um das im T2-Flair MR-Bild hyperintense Gebiet eingeplant. Verläßliche Vorhersagen des Tumorwachstums können dazu beitragen, die Strahlendosis noch besser auf gefährdete Regionen zu konzentrieren und gleichzeitig gesundes Gewebe zu schonen. Aktuelle Verfahren aus der Forschung nähern sich diesem Problem mit einer expliziten, generativen Modellierung des Wachstumsprozesses. Wir präsentieren ein alternatives, diskriminatives Verfahren. Mit Hilfe einer annotierten Datenbasis und überwachtem Lernen wird ein Wachstumsmodell trainiert und im nächsten Schritt auf ungesehene Daten angewendet. In allen 6 Testpatienten lieferte der Ansatz genauere Vorhersagen (DICE 0, 80±0, 09) als die bisherige Herangehensweise (DICE 0, 56 ± 0, 07).

Christian Weber, Michael Götz, Franciszek Binczyck, Joanna Polanska, Rafal Tarnawski, Barbara Bobek-Billewicz, Hans-Peter Meinzer, Bram Stieltjes, Klaus Maier-Hein

Classification of Confocal Laser Endomicroscopic Images of the Oral Cavity to Distinguish Pathological from Healthy Tissue

Confocal laser endomicroscopy is a recently introduced advanced imaging technique which enables microscopic imaging of the mucosa in-vivo. This technique has already been applied successfully during diagnosis of gastrointestinal diseases. Whereas for this purpose several computer aided diagnosis approaches exist, we present a classification system that is able to differentiate between healthy and pathological images of the oral cavity. Varying textural features of small rectangular regions are evaluated using random forests and support vector machines. Preliminary results reach up to 99.2% classification rate. This indicates that an automatic classification system to differentiate between healthy and pathological mucosa of the oral cavity is feasible.

Christian Jaremenko, Andreas Maier, Stefan Steidl, Joachim Hornegger, Nicolai Oetter, Christian Knipfer, Florian Stelzle, Helmut Neumann

Automatische Tumorsegmentierung mit spärlich annotierter Lernbasis

Die Erstellung von Trainingsdaten für lernbasierte Segmentierungsverfahren ist häufig sehr zeitaufwendig und fehleranfällig. Gleichzeitig muss die Lernbasis an die konkrete Bildgebung einer Klinik angepasst werden, was eine weite Verbreitung solcher automatischer Segmentierungsverfahren in der klinischen Routine verhindert. Wir schlagen daher ein Verfahren vor, welches durch die Verwendung eines Domain Adaption Ansatzes auf spärlichen, leicht anzufertigenden Segmentierungen trainiert werden kann. Wir validieren das vorgestellte System auf einem Kollektiv von 19 Patienten mit malignen Gliomen und zeigen, dass unser Ansatz die benötigte Annotierungszeit deutlich reduziert, während die Klassifikationsergebnisse gegenüber klassisch trainierten Segmentierungsans ätzen kaum beeinträchtigt werden. Der vorgestellte Ansatz erh öht die Attraktivität automatischer Segmentierungsverfahren für den klinischen Einsatz. Weiterhin lässt er die Erstellung umfangreicher Datenbanken mit großen Fallzahlen für unterschiedlichste Szenarien in greifbare Nähe rücken.

Michael Götz, Christian Weber, Franciszek Binczyck, Joanna Polanska, Rafal Tarnawski, Barbara Bobek-Billewicz, Hans-Peter Meinzer, Bram Stieltjes, Klaus Maier-Hein

3D Tensor Reconstruction in X-Ray Dark-Field Tomography

The First Phantom Result

X-ray dark-field imaging is a novel technique which provides complementary information on structural variation and density fluctuation. It allows to obtain object structures at micrometer scale and also contains information on the orientation of these structures. Since it can be acquired by a conventional X-ray imaging system, dark-field imaging has great potential for medical diagnosis. However, fully recovering 3D orientations in dark-field reconstruction still remains unexplored. In this paper, we propose an improved reconstruction method based on the zero-constrained dark-field reconstruction by Bayer et al. and a simplified principle axes transformation. A well-defined phantom containing representative 3D orientations is reconstructed in our experiment. On average, the structure orientations in the reconstructed volume differ from the ground truth by 9%. Within the boundaries of an object, the error drops to 6%. Application of this method in real diagnosis data can be expected in future.

Shiyang Hu, Christian Riess, Joachim Hornegger, Peter Fischer, Florian Bayer, Thomas Weber, Giesla Anton, Andreas Maier

Towards 3D Thyroid Imaging Using Robotic Mini Gamma Cameras

Thyroid imaging using radioactive tracers is a common task in clinics and is usually performed using 2D gamma cameras (scintigraphy). In this work we present a setup for 3D imaging of the thyroid using a mini gamma camera mounted on a robotic arm. Several images are acquired moving the mini gamma camera along a trajectory around the thyroid. Afterwards, a tomographic reconstruction computes a 3D SPECT-like image of the thyroid. First results are shown of a thyroid phantom using a conventional statistical reconstruction scheme (MLEM) and using a sparse regularization approach based on total variation.

Tobias Lasser, José Gardiazabal, Matthias Wieczorek, Philipp Matthies, Jakob Vogel, Benjamin Frisch, Nassir Navab

Investigation of Single Photon Emission Computed Tomography Acquired on Helical Trajectories

This study compares the quality of single photon emission computed tomography images obtained using step-and-shoot and helical trajectories. Monte Carlo simulations of an extended phantom on both trajectories were performed using a parallel hole collimator. Standard filtered-backprojection was used for reconstruction. Both trajectories collected data for the same amount of time. Corresponding to equivalent useful acquisition times, the background signal-to-noise ratios and sphere to background contrasts were roughly equivalent in both reconstructions. However, the helical trajectory requires 20% less true acquisition time due to the elimination of delay due to detector repositioning. Helical trajectories in SPECT can thus reduce overall acquisition time while having negligible effects on image quality.

Maximilian P. Oppelt, James C. Sanders, Andreas Maier

Blind Sparse Motion MRI with Linear Subpixel Interpolation

Vital and spontaneous motion causes major artifacts in MRI. In this paper a method is presented which reduces subpixel motion artifacts via computational post processing on a complete MR scan without additional data. On the compressed sparse MRI representation, translational subpixel motion is estimated iteratively from a fully sampled, but motion corrupted k-space, and motion free images are reconstructed by linear interpolation. Motion adjusted results are presented for the Shepp-Logan phantom and brainweb data.

Anita Möller, Marco Maaß, Alfred Mertins

Truncation Robust C-Arm CT Reconstruction for Dynamic Collimation Acquisition Schemes

Volume-of-interest (VOI) C-arm computed tomography (CT) imaging is a promising approach to acquire anatomical information in a pre-defined target volume at low dose, using both axial and trans-axial collimation. However, also the region outside the target volume, below referred to as peripheral region (PR), could contain some valuable information for image guidance. The potential use of a fast dynamically changing collimator would allow for new acquisition schemes, that acquire projection data in a way that allows for both a high-quality reconstruction of the diagnostic VOI and a low-quality reconstruction of the peripheral region, still at a low overall dose. In this paper, we present a novel reconstruction algorithm for an acquisition scheme that acquires a large portion of the projections in a collimated manner, while acquiring a small portion of the projections in a non-collimated manner. Experimental results indicate that few non-truncated projections can help to improve the image quality compared to a conventional VOI acquisition, while simultaneously providing valuable information about the peripheral region.

Thomas Kästner, Joachim Hornegger, Andreas Maier, Yan Xia, Sebastian Bauer

Backmatter

Weitere Informationen

Premium Partner

BranchenIndex Online

Die B2B-Firmensuche für Industrie und Wirtschaft: Kostenfrei in Firmenprofilen nach Lieferanten, Herstellern, Dienstleistern und Händlern recherchieren.

Whitepaper

- ANZEIGE -

Best Practices für die Mitarbeiter-Partizipation in der Produktentwicklung

Unternehmen haben das Innovationspotenzial der eigenen Mitarbeiter auch außerhalb der F&E-Abteilung erkannt. Viele Initiativen zur Partizipation scheitern in der Praxis jedoch häufig. Lesen Sie hier  - basierend auf einer qualitativ-explorativen Expertenstudie - mehr über die wesentlichen Problemfelder der mitarbeiterzentrierten Produktentwicklung und profitieren Sie von konkreten Handlungsempfehlungen aus der Praxis.
Jetzt gratis downloaden!

Bildnachweise