Angiographic images are routinely acquired for the medical diagnosis of arterial diseases. The digital images are in the form of three-dimensional (3D) volume, which consists of a stack of slices. Because of the lack of 3D information in those angiographic slices, radiologists generate maximum intensity project ion (MIP) images at different view angles to facilitate the diagnosis. This two-dimensional (2D) representation of the arterial morphology creates ambiguities in the assessment of position and size of arterial abnormalities. The motivation of this work stems from the clinical need for a 3D representation of the vascular structure....[ Read more ]

Angiographic images are routinely acquired for the medical diagnosis of arterial diseases. The digital images are in the form of three-dimensional (3D) volume, which consists of a stack of slices. Because of the lack of 3D information in those angiographic slices, radiologists generate maximum intensity project ion (MIP) images at different view angles to facilitate the diagnosis. This two-dimensional (2D) representation of the arterial morphology creates ambiguities in the assessment of position and size of arterial abnormalities. The motivation of this work stems from the clinical need for a 3D representation of the vascular structure.

In this work, two new segmentation techniques are presented for extracting 3D vessels from angiographic images and can be used complementarily. The first technique extracts 3D vessels based on MIP images using a unified platform to segment different kinds of angiograms; the second technique refines an initial 3D vascular segmentation based on orientation tensor used to analyze the local structure of the arterial morphology. Initial segmentation is then refined on a probabilistic framework based on local structure descriptions. We take the maximum a posteriori (MAP)-Markov random field (MRF) approach to solve the segmentation problem. The benefits of our segmentation techniques are twofold: (1) different kinds of angiograms are segmented with a unified algorithm; and (2) only intrinsic information (local structure) from arterial morphology is used to extract 3D vessels.

Experimental results on time-of-flight (TOF) magnetic resonance angiography (MRA), phase contrast (PC) MRA and 3D rotational angiography (RA) show that our algorithms outperform other methods in extracting vascular structures with low signal-to-noise ratios (SNR), e.g., small vessels and vessel edges.