Elsevier

Computers & Graphics

Volume 31, Issue 3, June 2007, Pages 493-500
Computers & Graphics

Technical Section
A hybrid pixel-based classification method for blood vessel segmentation and aneurysm detection on CTA

https://doi.org/10.1016/j.cag.2007.01.020Get rights and content

Abstract

In the present study, a hybrid semi-supervised pixel-based classification algorithm is proposed for the automatic segmentation of intracranial aneurysms in Computed Tomography Angiography images. The algorithm was designed to discriminate image pixels as belonging to one of the two classes: blood vessel and brain parenchyma. Its accuracy in vessel and aneurysm detection was compared with two other reliable methods that have already been applied in vessel segmentation applications: (a) an advanced and novel thresholding technique, namely the frequency histogram of connected elements (FHCE), and (b) the gradient vector flow snake. The comparison was performed by means of the segmentation matching factor (SMF) that expressed how precise and reproducible was the vessel and aneurysm segmentation result of each method against the manual segmentation of an experienced radiologist, who was considered as the gold standard. Results showed a superior SMF for the hybrid (SMF=88.4%) and snake (SMF=87.2%) methods compared to the FHCE (SMF=68.9%). The major advantage of the proposed hybrid method is that it requires no a priori knowledge of the topology of the vessels and no operator intervention, in contrast to the other methods examined. The hybrid method was efficient enough for use in 3D blood vessel reconstruction.

Introduction

An aneurysm is an abnormal bulging outward of an artery. Brain aneurysms (also called intracranial aneurysms (IAs)) are commonly located at the branching points of the major blood vessels at the base of the brain. IAs are usually discovered after they rupture, causing subarachnoid hemorrhage (SAH), i.e. bleeding in the subarachnoid space [1], [2]. A recent review [3] indicates a prevalence in the general population of up to 6%.

SAH is a serious condition with high morbidity and mortality. At 75% of the cases, SAHs are due to ruptured IAs, which usually originate from the major blood vessels of the cerebral arterial circle (circle of Willis) or from their branching arteries [4].

Detection of IAs has been traditionally performed using standard digital subtraction angiography (DSA) [5]. Computed Tomography Angiography (CTA) is a new non-invasive imaging modality that has recently started to be recognized as a rapid and accurate alternative to the standard DSA technique for brain aneurysms visualization [6], [7], [8], [9].

Brain aneurysm detection is of crucial importance, since it enables the quantification of a variety of crucial parameters (i.e. the width of the neck of the aneurysm, its orientation, and its relation to the parent vessel), that significantly affect treatment planning and surgical intervention [2], [6]. Thus, accurate detection and segmentation of brain blood vessels in CT angiograms is of major importance to the radiologists.

Although numerous studies and reviews have been devoted to the demanding task of vessel segmentation [10], [11], [12], [13], [14], from the technical aspect, there is no general technique that may be effectively applied to all modalities [12]. Regarding CTA, most previous studies have used inbuilt CT-software to segment aneurysms [5], [8], others have employed commercial software to outline aneurysms [6], [9], and few have experimented with own developed software for segmentation [10], [13], [14]. However, most segmentation techniques require a priori knowledge and/or operator intervention. Snakes (active contours) have been shown to be among the most promising techniques [10], [14], [15]. Another promising segmentation method is the pixel-based classification, employing supervised or unsupervised classifiers [16], [17], [18], that require no a priori information and, to our knowledge, have found no application in vessel segmentation.

In the present study, a hybrid semi-supervised pixel-based segmentation algorithm for the segmentation of aneurysms on CTA is proposed and it is compared against two other novel and reliable techniques that have been previously tested in vessel segmentation applications: an advanced thresholding technique, using connected elements, that has been employed in vessel segmentation on DSA images, and an advanced snake method [19], that has been previously used in aneurysm segmentation on CTA images.

Section snippets

Material and methods

Eleven cases of patients with IA were examined. From each case, a Digital Imaging and Communications in Medicine (DICOM) dataset of CTA brain images (Siemens Somaton Plus 4, Siemens AG, Erlangen, Germany) was acquired from the Department of Radiology of the University Hospital of Patras, Greece. Dataset comprised 924 CTA images in total. CTA data slices were interpreted, using a typical window width (150 HU) and window center (100 HU), by an expert radiologist (T.P.). As the basic interest was

Results

Fig. 3 illustrates the output regions of the k-means algorithm for vessels and brain parenchyma. From these regions, features were calculated to represent the vessels and brain parenchyma class, respectively.

The features mean value (derived from the ROIs histogram), inverse difference moment, and difference entropy (co-occurrence matrix features, see Appendix B) comprised the best feature vector combination, that optimized the performance of the PNN, in classifying aneurysm-vessel from

Discussion

In the present study, a hybrid semi-supervised pixel-based classification method was introduced for segmenting brain blood vessels and, thus, possible attached aneurysms. This hybrid algorithm comprised an unsupervised clustering algorithm (k-means) that led a supervised pixel-based segmentation algorithm, without any a priori knowledge of the shape and topology of the aneurysm to be segmented.

The proposed hybrid segmentation method was tested against two state-of-art segmentation algorithms,

Acknowledgment

We would like to thank the Greek State Scholarships Foundation (I.K.Y.) for funding the above work.

References (29)

  • B.F. Tomandl et al.

    CT angiography of intracranial aneurysms: a focus on postprocessing

    Radiographics

    (2004)
  • S. Chawla

    Advances in multidetector computed tomography: applications in neuroradiology

    Journal of Computer Assisted Tomography

    (2004)
  • M. Kangansiemi et al.

    Detection of intracranial aneurysms with two-dimensional and three-dimensional multislice helical computed tomographic angiography

    Neurosurgery

    (2004)
  • J.P. Villablanca et al.

    Detection and characterization of very small cerebral aneurysms by using 2D and 3D helical CT angiography

    American Journal of Neuroradiology

    (2002)
  • Cited by (9)

    • Multi-branched cerebrovascular segmentation based on phase-field and likelihood model

      2014, Computers and Graphics (Pergamon)
      Citation Excerpt :

      Vascular imaging plays an important role in tackling vascular diseases. In particular, vessel segmentation is a fundamental task in the diagnosis and treatment planning of many different pathological symptoms including arteriovenous malformations (AVMs), aneurysm, thrombosis and cardiac disease [1]. Therefore, accurately segmenting blood vessels, especially fine vessels on angiograms of different modalities, has gained much research attention [2].

    • Aneurysm identification by analysis of the blood-vessel skeleton

      2013, Computer Methods and Programs in Biomedicine
      Citation Excerpt :

      Not many research papers proposing automation of this process have been published [6–9]. Fully automatic or semi-automatic detection of aneurysms is based mainly on the segmentation of CT scans (or angiography images) followed by a probabilistic classification of the segmented regions [6,8] exploiting the fact that voxels corresponding to the aneurysm have a different intensity than voxels of blood vessels or the surrounding mass [9]. The accuracy of these methods varies, in accordance with the screening resolution and the kind of segmentation algorithm used.

    View all citing articles on Scopus
    View full text