Whole heart segmentation from cardiac CT scans is a prerequisite for many clinical applications, but manual delineation is a tedious task and subject to both intra- and inter-observer variation. Automating the segmentation process has thus become an increasingly popular task in the field of image analysis, and is generally solved by either using 3D methods, considering the image volume as a whole, or 2D methods, segmenting each slice independently. In the field of deep learning, there are significant limitations regarding 3D networks, including the need for more training examples and GPU memory. The need for GPU memory is usually solved by down sampling the input images, thus losing important information, which is not a necessary sacrifice when employing 2D networks. It would therefore be relevant to exploit the benefits of 2D networks in a configuration, where spatial information across slices is kept, as when employing 3D networks. The proposed method performs multiclass segmentation of cardiac CT scans utilizing 2D convolutional neural networks with a multi-planar approach. Furthermore, spatial propagation is included in the network structure, to ensure spatial consistency through each image volume. The approach keeps the computational assets of 2D methods while addressing 3D issues regarding spatial context. The pipeline is structured in a two-step approach, in which the first step detects the location of the heart and crops a region of interest, and the second step performs multi-class segmentation of the heart structures. The pipeline demonstrated promising results on the MICCAI 2017 Multi-Modality Whole Heart Segmentation challenge data.
In this paper we present a novel sensing system, robust Near-infrared Structured Light Scanning (NIRSL) for three-dimensional human model scanning application. Human model scanning due to its nature of various hair and dress appearance and body motion has long been a challenging task. Previous structured light scanning methods typically emitted visible coded light patterns onto static and opaque objects to establish correspondence between a projector and a camera for triangulation. In the success of these methods rely on scanning objects with proper reflective surface for visible light, such as plaster, light colored cloth. Whereas for human model scanning application, conventional methods suffer from low signal to noise ratio caused by low contrast of visible light over the human body. The proposed robust NIRSL, as implemented with the near infrared light, is capable of recovering those dark surfaces, such as hair, dark jeans and black shoes under visible illumination. Moreover, successful structured light scan relies on the assumption that the subject is static during scanning. Due to the nature of body motion, it is very time sensitive to keep this assumption in the case of human model scan. The proposed sensing system, by utilizing the new near-infrared capable high speed LightCrafter DLP projector, is robust to motion, provides accurate and high resolution three-dimensional point cloud, making our system more efficient and robust for human model reconstruction. Experimental results demonstrate that our system is effective and efficient to scan real human models with various dark hair, jeans and shoes, robust to human body motion and produces accurate and high resolution 3D point cloud.
Twenty-three Taiwanese infants with unilateral cleft lip and palate (UCLP) were CT-scanned before lip repair at the age
of 3 months, and again after lip repair at the age of 12 months. In order to evaluate the surgical result, detailed point
correspondence between pre- and post-surgical images was needed. We have previously demonstrated that non-rigid
registration using B-splines is able to provide automated determination of point correspondences in populations of
infants without cleft lip. However, this type of registration fails when applied to the task of determining the complex
deformation from before to after lip closure in infants with UCLP. The purpose of the present work was to show that use
of prior information about typical deformations due to lip closure, through the construction of a knowledge-based atlas of
deformations, could overcome the problem. Initially, mean volumes (atlases) for the pre- and post-surgical populations,
respectively, were automatically constructed by non-rigid registration. An expert placed corresponding landmarks in the
cleft area in the two atlases; this provided prior information used to build a knowledge-based deformation atlas. We
model the change from pre- to post-surgery using thin-plate spline warping. The registration results are convincing and
represent a first move towards an automatic registration method for dealing with difficult deformations due to this type
of surgery.
Patient motion during scanning deteriorates image quality, especially for high resolution PET scanners. A new proposal
for a 3D head tracking system for motion correction in high resolution PET brain imaging is set up and demonstrated. A
prototype tracking system based on structured light with a DLP projector and a CCD camera is set up on a model of the
High Resolution Research Tomograph (HRRT). Methods to reconstruct 3D point clouds of simple surfaces based on
phase-shifting interferometry (PSI) are demonstrated. The projector and camera are calibrated using a simple stereo
vision procedure where the projector is treated as a camera. Additionally, the surface reconstructions are corrected for the
non-linear projector output prior to image capture. The results are convincing and a first step toward a fully automated
tracking system for measuring head motions in PET imaging.
We address the problem of intra-subject registration for change detection. The goal is to separate stationary and changing
subsets to be able to robustly perform rigid registration on the stationary subsets and thus improve the subsequent change
detection. An iterative approach using a hybrid of parametric and non-parametric statistics is presented. The method
uses non-parametric clustering and large scale hypothesis testing with estimation of the empirical null hypothesis. The
method is successfully applied to 3D surface scans of human ear impressions containing true changes as well as data with
synthesized changes. It is shown that the method improves registration and is capable of reducing the difference between
registration using different norms.
This work describes a non-rigid registration method for open 2D manifold embedded in 3D Euclidian space. The
method is based on difference of distance maps and grid based warps interpolated by splines constrained in such
a way that the deformation field is diffeomorphic. We then create a dense surface to surface correspondence using
angle weighted normals and ray tracing. The implementation using a derivation of the inverse compositional
algorithm for optimization of computational speed is described. The results are evaluated as a shape model
showing the principal modes of variation.
We evaluate a novel method for fully automated rigid registration of 2D manifolds in 3D space based on distance
maps, the Gibbs sampler and Iterated Conditional Modes (ICM). The method is tested against the ICP considered
as the gold standard for automated rigid registration. Furthermore, the influence of different norms and sampling
point densities is evaluated. The performance of the two methods has been evaluated on data consisting of 178
scanned ear impressions taken from the right ear. To quantify the difference of the two methods we calculate
the registration cost and the mean point to point distance. T-test for common mean are used to determine
the performance of the two methods (supported by a Wilcoxon signed rank test). The performance influence of
sampling density, sampling quantity, and norms is analyzed using a similar method.
In this paper it is described how to build a statistical shape model using a training set with a sparse of landmarks. A well defined model mesh is selected and fitted to all shapes in the training set using thin plate spline warping. This is followed by a projection of the points of the warped model mesh to the target shapes. When this is done by a nearest neighbour projection it can result in folds and inhomogeneities in the correspondence vector field. The novelty in this paper is the use and extension of a Markov random field regularisation of the correspondence field. The correspondence field is regarded as a collection of random variables, and using the Hammersley-Clifford theorem it is proved that it can be treated as a Markov Random Field. The problem of finding the optimal correspondence field is cast into a Bayesian framework for Markov Random Field restoration, where the prior distribution is a smoothness term and the observation model is the curvature of the shapes. The Markov Random Field is optimised using a combination of Gibbs sampling and the Metropolis-Hasting algorithm. The parameters of the model are found using a leave-one-out approach. The method leads to a generative model that produces highly homogeneous polygonised shapes with improved reconstruction capabilities of the training data. Furthermore, the method leads to an overall reduction in the total variance of the resulting point distribution model. The method is demonstrated on a set of human ear canals extracted from 3D-laser scans.
Today the design of custom completely-in-the-canal hearing aids is a
manual process and therefore there is a variation in the quality of
the finished hearing aids. Especially the placement of the so-called
faceplate on the hearing aid strongly influences the size and shape
of the hearing aid. Since the future hearing aid production will be
less manual there is a need for algorithms that mimic the
craftsmanship of skilled operators. In this paper it is described
how a statistical shape model of the ear canal can be used to
predict the placement of the faceplate on a hearing aid made for a
given ear canal. The shape model is a point distribution model built
using a training set of shapes with manually placed landmarks. An
interpolation method is used to generate dense landmark
correspondence over the training set prior to building the shape
model. Faceplates have also been placed on the training shapes by a
skilled operator. These faceplate planes are aligned to the average
shape from the shape model and an average faceplate plane is
calculated. Given a surface representation of a new ear canal, the
shape model is fitted using a combination of the iterative closest
point algorithm and the active shape model approach. The average
faceplate from the training set can now be placed on the new ear
canal using the position of the fitted shape model. A leave-one-out
study shows that the algorithm is able to produce results comparable
to a human operator.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
INSTITUTIONAL Select your institution to access the SPIE Digital Library.
PERSONAL Sign in with your SPIE account to access your personal subscriptions or to use specific features such as save to my library, sign up for alerts, save searches, etc.