Cerebrovascular Segmentation via Vessel Oriented Filtering Network Zhanqiang Guo1 Yao Luan Jianjiang Feng1B Wangsheng Lu Yin Yin

2025-04-30 0 0 2.5MB 10 页 10玖币

侵权投诉

Cerebrovascular Segmentation via Vessel

Oriented Filtering Network

Zhanqiang Guo1, Yao Luan, Jianjiang Feng1(B), Wangsheng Lu, Yin Yin,

Guangming Yang, and Jie Zhou

Department of Automation, Tsinghua University, Beijing, China

Abstract. Accurate cerebrovascular segmentation from Magnetic Res-

onance Angiography (MRA) and Computed Tomography Angiography

(CTA) is of great signiﬁcance in diagnosis and treatment of cerebrovas-

cular pathology. Due to the complexity and topology variability of blood

vessels, complete and accurate segmentation of vascular network is still

a challenge. In this paper, we proposed a Vessel Oriented Filtering Net-

work (VOF-Net) which embeds domain knowledge into the convolutional

neural network. We design oriented ﬁlters for blood vessels according to

vessel orientation ﬁeld, which is obtained by orientation estimation net-

work. Features extracted by oriented ﬁltering are injected into segmenta-

tion network, so as to make use of the prior information that the blood

vessels are slender and curved tubular structure. Experimental results

on datasets of CTA and MRA show that the proposed method is eﬀec-

tive for vessel segmentation, and embedding the speciﬁc vascular ﬁlter

improves the segmentation performance.

Keywords: Cerebrovascular ·Vessel Segmentation ·Magnetic Reso-

nance Angiography ·Computed Tomography Angiography ·Oriented

Filter Network

1 Introduction

Cerebrovascular disease is one of the leading causes of death and disability world-

wide [1]. Magnetic Resonance Angiography (MRA) and Computed Tomography

Angiography (CTA) are often applied to imaging the cerebrovascular system in

clinical diagnosis. Accurate segmentation of cerebral vasculature from MRA and

CTA images is the ﬁrst step for many clinical applications. However, segment-

ing cerebral vessels is very challenging due to various diﬃculties, such as the

complex cerebral vascular structure, high degree of anatomical variation, small

vessel size, poor vessel contrast, and data sparseness [2,3].

In recent decades, a number of automatic (or semi-automatic) vascular seg-

mentation methods have been proposed [4,5]. Existing cerebrovascular segmen-

tation algorithms can be coarsely divided into two categories: traditional algo-

rithms and deep learning based algorithms. The key of the ﬁrst type of method

is to artiﬁcially formulate some rules to extract blood vessel features, which

usually take into account the fact that the vessels are thin and slender tubu-

lar structure. For example, many segmentation methods are based on features

arXiv:2210.08868v1 [eess.IV] 17 Oct 2022

2 Zhanqiang Guo et al.

computed from hessian matrix [6,7]. Combined with the features extracted by

the rules formulated artiﬁcially, some other traditional methods are also used for

vessel segmentation. Yang et al. [8] presented a geodesic activate contour model

to segment blood vessels, which adaptively conﬁgures parameters. Zhao et al.

[9] proposed a likelihood model based on level set, achieving good results for

segmentation of vascular network. Lv et al. [10] proposed a parallel algorithm

based on heterogeneous Markov random ﬁeld to segment vessels. However, these

methods depend heavily on domain knowledge which may be inaccurate, and

thus are not robust to image quality.

Recently, deep learning methods are widely used in the ﬁeld of medical image

processing. The segmentation algorithms based on convolutional neural network

have been shown to produce state of the art results in various medical segmenta-

tion tasks [11,12,13,14]. One of the most notable methods is 3DUnet [15]. Similar

to it, Fausto et al. [16] used convolution instead of max-pooling and residual con-

nections to improve performance. Inspired by the inception model [17], Pedro et

al. [18] proposed Uception for vascular segmentation, which increases the net-

work size to have a better representation. To reduce computational burden and

memory consumption in 3D segmentation, 2D orthogonal cross-hair ﬁlters were

formulated in DeepVesselNet [19]. Zhang et al. [20] proposed RENet to remove

redundant features and retain edge information in shallow features, and reported

good results in segmentation of cerebral vessels. Based on Multiple-U-Net, Guo

et al. [21] proposed a novel network for vessel segmentation. Aimed for CTA

images, Fu et al. [22] proposed a framework based 3D convolutional neural net-

work for segmenting cerebrovascular networks. However, these methods fail to

explicitly consider the peculiarities of the vascular structure, namely, blood ves-

sels are curved and elongated tubular structure. It is diﬃcult and unpredictable

to rely solely on network training to learn these characteristics of vessels from

training data, especially when the size of training set is limited, which is typical

in medical image analysis.

In response to the problems mentioned above, we propose a Vessel Oriented

Filtering Network (VOF-Net) that combines domain knowledge and deep learn-

ing method. First, we propose a vessel orientation estimation network to calcu-

late the orientation ﬁeld of blood vessels, through which vessel oriented ﬁlters are

designed. Then, we feed the extracted vessel features to a 3DUnet based vessel

segmentation network, which is trained to combine information from features

extracted by oriented ﬁlters. Experiments on CTA and MRA datasets show that

our proposed method yields better result than other popular vascular segmen-

tation methods.

2 Methodology

Our proposed system, as shown in Fig. 1, consists of three modules: Orientation

Estimation Network, Oriented Filtering Module, and Segmentation Network.

文档加载中……请稍候！
如果长时间未打开，您也可以点击刷新试试。

下载文档到电脑，查找使用更方便

10 玖币 0人已下载

立即下载

摘要：

CerebrovascularSegmentationviaVesselOrientedFilteringNetworkZhanqiangGuo1,YaoLuan,JianjiangFeng1(B),WangshengLu,YinYin,GuangmingYang,andJieZhouDepartmentofAutomation,TsinghuaUniversity,Beijing,ChinaAbstract.AccuratecerebrovascularsegmentationfromMagneticRes-onanceAngiography(MRA)andComputedTomograph...

展开>> 收起<<

Cerebrovascular Segmentation via Vessel Oriented Filtering Network Zhanqiang Guo1 Yao Luan Jianjiang Feng1B Wangsheng Lu Yin Yin.pdf

共10页,预览2页

还剩页未读，继续阅读

声明：本站为文档C2C交易模式，即用户上传的文档直接被用户下载，本站只是中间服务平台，本站所有文档下载所得的收益归上传人(含作者)所有。玖贝云文库仅提供信息存储空间，仅对用户上传内容的表现方式做保护处理，对上载内容本身不做任何修改或编辑。若文档所含内容侵犯了您的版权或隐私，请立即通知玖贝云文库，我们立即给予删除！

Cerebrovascular Segmentation via Vessel Oriented Filtering Network Zhanqiang Guo1 Yao Luan Jianjiang Feng1B Wangsheng Lu Yin Yin

相关推荐

开通VIP享超值会员特权

作者详情

相关内容

热门标签

举报选择: