Logo des Repositoriums

Auflistung nach Autor:in "Reisert, Marco"

1 - 3 von 3
  • Konferenzbeitrag
    Dense rotation invariant brain pyramids for automated human brain parcellation
    (INFORMATIK 2011 – Informatik schafft Communities, 2011) Skibbe, Henrik; Reisert, Marco
    The automatic parcellation of the human brain based on MR imaging is in several areas of high interest. In particular, identifying corresponding brain areas between different subjects is an indispensable prerequisite for any group analysis. But also, simple segmentations into different tissue types is an important preprocessing step. We present a generic framework for describing and automatically parcellating high angular resolution diffusion-weighted magnetic-resonance images (HARDI) of the human brain. Based on an initial training step our approach is capable to segment the images into coarse parcellations or detailed fine grain regions of interest. In contrast to existing model-free methods [SSK+09] we are not only using the raw measurements at each position, but we are also including neighboring measurements in a rotation invariant way.
  • Konferenzbeitrag
    Diffusion propagator imaging by model-driven regularization
    (INFORMATIK 2011 – Informatik schafft Communities, 2011) Reisert, Marco; Kiselev, Valerij G.
    Diffusion-weighted magnetic resonance imaging is able to non-invasively visualize the fibrous structure of the human brain white matter. The robust and accurate estimation of the ensemble average diffusion propagator (EAP), based on diffusionsensitized magnetic resonance images, is an important preprocessing step for tractography algorithms or any other derived statistical analysis. In this work, we propose a new regularization strategy for EAP estimation that bridges the gap between modelbased and model-free approaches. The idea is to use a Gaussian prior density which is especially designed for the diffusion signal in the human brain. Therefore, we propose to compute covariance statistics over a family of functions that are typical for human brain white matter. As the considered functions and the physically observed EAPs are usually smooth and local the Gauss-Laguerre basis system is used for realization. With this methodology it is possible to estimate the whole 3D EAP from a single qshell measurement. In comparison to usual extrapolation strategies our approach is linear in the measured signal which makes it more robust to noise and partial volume effects. We will show this in synthetic and in-vivo experiments.
  • Textdokument
    Equivariante Kerne in der Mustererkennung
    (Ausgezeichnete Informatikdissertationen 2008, 2009) Reisert, Marco
    Einer der wichtigsten Einflussfaktoren im Entwurf eines Mustererkennungssystems stellt a-priori Wissen dar. A-priori Wissen liefert Information welche über die reinen Trainingsdaten hinausgeht. Eine wichtige Untergruppe ist das sogenannte Transformationswissen. Man weiss, welche Gruppe von Transformationen die Bedeutung der betrachteten Muster unverändert lässt. In der Literatur existieren viele ad-hoc Ansätze wie solches Wissen miteinbezogen werden kann. In dieser Dissertation wurde ein wohlfundiertes mathematisches Rahmenwerk ausgearbeitet, welches klärt, in welcher Weise sich solches Wissen optimal in ein System integrieren lässt. Aus praktischer Perspektive wird eine Methode vorgestellt, wie sich explizit Equivarianzbedingungen in kernbasierte und parametrische Modellbeschreibungen integrieren lassen. In diesem Artikel werden die Grundbegriffe und Hauptresultate präsentiert und ein Anwendungsbeispiel zur rotationsinvarianten Objektdetektion besprochen.