X-ray coronary angiography (XCA) is a minimally invasive and common method for the diagnosis of coronary artery diseases. Fast and accurate segmentation vessel structure from XCA sequence is significant to assist doctors’ treatment. However, vessel segmentation is relatively challenging due to low contrast, presence of noise and overlaps from adjacent tissues. In this paper, we develop a novel network model based on conventional u-net architecture. To exploit the rich temporal-spatial information in XCA and provide consistent contexts for vessel inference, we take multi adjacent frames from XCA as input and adopt several 3D convolutional layers in the encode stage to extract temporal-spatial features representation. In skip connection layers, salient mechanism is utilized to adaptively filter out the noise from raw temporal-spatial features and stress vessel feature components. After that, the refined 3D temporal-spatial features are fused along with the temporal axis to generate 2D feature representations. These 2D feature representations are then passed to corresponding 2D upsampling layers in the decode stage. Experimental results verify the feasibility of salient mechanism application. Besides, extensive experiments demonstrate our superior performance over other state-of-the-art methods.
In ultrasound despeckling, it is essential to remove speckle noise with satisfactory feature preservation for better diagnosis and analysis in many applications. This paper proposes an adaptive fractional-order differentiation filter guided by feature asymmetry for feature-preserving ultrasound despeckling. Since fractional anisotropic diffusion (FAD) filter performs well in smooth regions while fractional total variation (FTV) filter works better near features, our framework combines the FAD filter and FTV filter to maintain their advantages. Moreover, the feature significance calculated by using feature asymmetry is integrated into the diffusion coefficient of the FAD filter to protect low contrast features. Finally, rather than adopting one fixed fractional order, the proposed filter adaptively assigns fractional order on the basis of the feature significance to further preserve features. Experiments on synthetic and clinic ultrasound images demonstrate that the proposed filter performs better in both speckle reduction and feature preservation compared with other state-of-the-art ultrasound speckle reduction filters.
Image registration is a process of creating correspondence between a pair of images. In some situations, the physical one-to-one correspondence may not exist due to the presence of "outlier" objects (called gross outliers) that appear in one
image but not the other. In this paper, a novel robust method is presented to address the problem of tumor-like gross
outliers in non-rigid image registration. First, two salient point sets are extracted from the two images to be registered,
and classified by means of clustering analysis which is based on Gaussian mixture models and expectation-maximization
(EM) algorithm. Then by means of joint saliency map that represents the joint salient regions of the overlapping volume
of the two images, the regions including tumor-like gross outliers could be automatically recognized. After screening out
of salient points and elimination of outlier points, some stable control points that well represent the corresponding
structures within the joint salient regions of the two images could be obtained. By iteratively finding correspondences
between the control points in the joint salient regions, the smooth deformation field is approximated based on radial basis
functions (RBFs) with compact support until the convergence to the steady-state solution is achieved. Experimental
results show that the proposed method is able to recover local deformation caused by tumor resection in brain.
A novel non-rigid registration algorithm within multi-resolution block matching framework is presented for accurate and
robust image registration in the presence of incomplete image information. After getting the deformation field computed
from block-matching, we introduce robust and structure-adaptive normalized convolution in spatial regularization of
deformation field. Unlike traditional framework of normalized convolution, in which the local deformation is modified
through a projection onto a subspace, however, the applicability function of structure-adaptive normalized convolution
based on an anisotropic Gaussian kernel is adapted to local linear or edge structures in the images to be registered. This
leads to more samples of regions of homogeneity being gathered for the regularization of deformation field, which can
reduce deformation diffusion across discontinuities. A robust signal certainty is also adapted to each displacement vector
in the deformation field to measure its accuracy. The results show that the method is sufficiently accurate and robust to
incomplete image information for multi-temporal non-rigid image registration.
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.