Search

Image registration is an essential technique to obtain point correspondences between anatomical structures from different images. Conventional non-rigid registration methods assume a continuous and smooth deformation field throughout the image. However, the deformation field at the interface of different organs is not necessarily continuous, since the organs may slide over or separate from each other. Therefore, imposing continuity and smoothness ubiquitously would lead to artifacts and increased errors near the discontinuity interface. In computational mechanics, the eXtended Finite Element Method (XFEM) was introduced to handle discontinuities without using computational meshes that conform to the discontinuity geometry. Instead, the interpolation bases themselves were enriched with discontinuous functional terms. Borrowing this concept, we propose a multiresolution eXtented Free-Form Deformation (XFFD) framework that seamlessly integrates within and extends the standard Free-Form Deformation (FFD) approach. Discontinuities are incorporated by enriching the B-spline basis functions coupled with extra degrees of freedom, which are only introduced near the discontinuity interface. In contrast with most previous methods, restricted to sliding motion, no ad hoc penalties or constraints are introduced to reduce gaps and overlaps. This allows XFFD to describe more general discontinuous motions. In addition, we integrate XFFD into a rigorously formulated multiresolution framework by introducing an exact parameter upsampling method. The proposed method has been evaluated in two publicly available datasets: 4D pulmonary CT images from the DIR-Lab dataset and 4D CT liver datasets. The XFFD achieved a Target Registration Error (TRE) of 1.17 ± 0.85 mm in the DIR-lab dataset and 1.94 ± 1.01 mm in the liver dataset, which significantly improves on the performance of the state-of-the-art methods handling discontinuities. [ABSTRACT FROM AUTHOR]

Purpose: The primary goal of this article is to achieve an automatic and objective method to compute the Pfirrmann's degeneration grade of intervertebral discs (IVD) from MRI. This grading system is used in the diagnosis and management of patients with low back pain (LBP). In addition, biomechanical models, which are employed to assess the treatment on patients with LBP, require this grading value to compute proper material properties.Materials and Methods: T2-weighted MR images of 48 patients were employed in this work. The 240 lumbar IVDs were divided into a training set (140) and a testing set (100). Three experts manually classified the whole set of IVDs using the Pfirrmann's grading system and the ground truth was selected as the most voted value among them. The developed method employs active contour models to delineate the boundaries of the IVD. Subsequently, the classification is achieved using a trained Neural Network (NN) with eight designed features that contain shape and intensity information of the IVDs.Results: The classification method was evaluated using the testing set, resulting in a mean specificity (95.5 %) and sensitivity (87.3 %) comparable to those of every expert with respect to the ground truth.Conclusions: Our results show that the automatic method and humans perform equally well in terms of the classification accuracy. However, human annotations have inherent inter- and intra-observer variabilities, which lead to inconsistent assessments. In contrast, the proposed automatic method is objective, being only dependent on the input MRI. [ABSTRACT FROM AUTHOR]

Statistical shape models (SSM) are used to introduce shape priors in the segmentation of medical images. However, such models require large training datasets in the case of multi-object structures, since it is required to obtain not only the individual shape variations but also the relative position and orientation among objects. A solution to overcome this limitation is to model each individual shape independently. However, this approach does not take into account the relative position, orientations and shapes among the parts of an articulated object, which may result in unrealistic geometries, such as with object overlaps. In this article, we propose a new Statistical Model, the Statistical Interspace Model (SIM), which provides information about the interaction of all the individual structures by modeling the interspace between them. The SIM is described using relative position vectors between pair of points that belong to different objects that are facing each other. These vectors are divided into their magnitude and direction, each of these groups modeled as independent manifolds. The SIM was included in a segmentation framework that contains an SSM per individual object. This framework was tested using three distinct types of datasets of CT images of the spine. Results show that the SIM completely eliminated the inter-process overlap while improving the segmentation accuracy. [ABSTRACT FROM AUTHOR]