OCTRexpert：AFeature-based3DRegistrationMethodforRetinalOCTImages资源-CSDN文库

56 浏览量 2021-02-11 21:25:29 上传评论收藏 1.35MB PDF 举报

### OCTRexpert：一种基于特征的三维注册方法用于视网膜OCT图像 #### 概述 OCTRexpert是一种创新的、基于特征的三维（3D）注册算法，旨在为视网膜光学相干断层成像（Optical Coherence Tomography, OCT）提供更精确的配准解决方案。此方法克服了现有技术在处理复杂病理情况下视网膜OCT图像时的局限性，特别是在纵向数据比较方面。OCTRexpert能够实现对正常和严重病理患者视网膜OCT图像的精确配准。 #### 研究背景与意义医学图像配准是生物医学研究中的关键技术之一，广泛应用于纵向和横断面数据分析、疾病进展的定量监测以及计算机辅助诊断和治疗。然而，在视网膜OCT图像领域，能够实现精确且定量对比的变形配准方法尚未得到充分发展。视网膜OCT图像的高分辨率特性使其成为眼科临床实践中不可或缺的工具，但这也增加了图像配准的挑战性。OCTRexpert的提出填补了这一领域的空白，并且能够有效应用于不同健康状况患者的纵向OCT图像分析。 #### 主要贡献 OCTRexpert的主要贡献包括： 1. **全三维注册**：这是首个专为视网膜OCT图像设计的全三维配准算法，适用于纵向图像集，不仅限于正常情况下的图像，也适用于严重病理状态下的图像。 2. **预处理方法**：该方法首先去除眼动伪影，为后续的配准步骤奠定基础。 3. **设计-检测-变形策略**： - 在设计阶段，为每个体素设计一组特征。 - 在检测阶段，选择活跃体素并建立主体和模板图像之间的点对点对应关系。 - 在变形阶段，根据多分辨率下检测到的对应关系，采用分层方式对图像进行变形。 4. **评估结果**：实验结果表明，OCTRexpert相比其他配准方法在Dice相似系数和平均无符号表面误差方面具有显著优势。 #### 技术细节 1. **特征设计**：为了提高配准的准确性，OCTRexpert为每个体素设计了一组特征，这些特征能够反映局部结构的信息，从而增强配准过程中的鲁棒性和准确性。 2. **活跃体素的选择**：通过选择活跃体素，可以减少计算量，同时确保关键结构的准确对应。 3. **多分辨率变形**：通过在多个分辨率层次上执行变形操作，可以有效地处理大范围的变形，同时保持细节的准确性。 #### 应用案例 OCTRexpert已经在一个包含来自20名健康个体和4名被诊断为严重脉络膜新生血管（Choroidal Neovascularization, CNV）患者的纵向OCT图像的数据集上进行了评估。实验结果显示，该方法在定量指标上取得了显著改善，证明了其在实际应用中的有效性和优越性。 #### 结论 OCTRexpert作为首个专为视网膜OCT图像设计的全三维配准算法，通过创新的设计-检测-变形策略，成功地提高了配准的精度和可靠性。该方法不仅适用于正常情况下的图像，也适用于严重病理状态下的图像，为纵向数据的精确比较提供了强有力的支持。此外，通过国家自然科学基金等项目的资助，OCTRexpert的研发工作得到了有力支持，这标志着中国在视网膜OCT图像配准领域的研究已达到国际先进水平。 OCTRexpert作为一种先进的配准算法，对于推动视网膜疾病的早期诊断和治疗具有重要意义。

资源推荐

资源详情

资源评论



Abstract—Medical image registration can be used for studying

longitudinal and cross-sectional data, quantitatively monitoring

disease progression and guiding computer assisted diagnosis and

treatments. However, deformable registration which enables

more precise and quantitative comparison has not been well

developed for retinal optical coherence tomography (OCT)

images. This paper proposes a new 3D registration approach for

retinal OCT data called OCTRexpert. To the best of our

knowledge, the proposed algorithm is the first full 3D registration

approach for retinal OCT images which can be applied to

longitudinal OCT images for both normal and serious

pathological subjects. In this approach, a pre-processing method

is first performed to remove eye motion artifact and then a novel

design-detection-deformation strategy is applied for the

registration. In the design step, a couple of features are designed

for each voxel in the image. In the detection step, active voxels are

selected and the point-to-point correspondences between the

subject and template images are established. In the deformation

step, the image is hierarchically deformed according to the

detected correspondences in multi-resolution. The proposed

method is evaluated on a dataset with longitudinal OCT images

from 20 healthy subjects and 4 subjects diagnosed with serious

Choroidal Neovascularization (CNV). Experimental results show

that the proposed registration algorithm consistently yields

statistically significant improvements in both Dice similarity

coefficient and the average unsigned surface error compared with

the other registration methods.

Index Terms—Image registration, Optical coherence

tomography (OCT), retinal image.

This work was supported by the National key R & D plan "key scientific

issues of transformative technology" under 2018YFA0701700 ， National

Nature Science Foundation of China for Excellent Young Scholars under

61622114，in part by the National Natural Science Foundation of China

(NSFC) under Grant 61601317 and 61601207, in part by the Natural Science

Foundation for colleges and universities of the Jiangsu Province under Grant

17KJB510015 , in part by the Changzhou Social Development Foundation

under Grant CE20165021, in part by Collaborative Innovation Center of IoT

Industrialization and Intelligent Production, Minjiang University under Grant

IIC1702. (Corresponding author: Xinjian Chen.)

L. Pan is with the School of Electronic and Information Engineering,

Soochow University, China. She is also with the School of Electrical and

Information Engineering, Jiangsu University of Technology (e-mail:

20154028001@stu.suda.edu.cn). F. Shi and Kai Yu are with the School of

Electronic and Information Engineering, Soochow University, China. They are

also with the Collaborative Innovation Center of IoT Industrialization and

Intelligent Production, Minjiang University. L.Duan and J. Zheng is with

Suzhou Institute of Biomedical Engineering and Technology, Chinese

Academy of Sciences, Suzhou, China. X. Chen and D. Xiang are with the

School of Electronic and Information Engineering, Soochow University, China,

and also with the State Key Laboratory of Radiation Medicine and Protection,

School of Radiation Medicine and Protection, Soochow University, Suzhou

China. (e-mail: xjchen@suda.edu.cn,).

I. INTRODUCTION

EDICAL image registration has attracted more and more

attention due to its valuable applications in clinical

studies [1]. In particular, registration can be used for studying

longitudinal and cross-sectional data, quantitatively monitoring

disease progression and guiding computer assisted diagnosis

and treatments. Registration is a fundamental task in medical

image processing used to match multiple images taken from

different viewpoints, different sensors or different time points.

In the past decades, many registration techniques have been

developed for various types of data and applications. However,

spectral-domain optical coherence tomography (SD-OCT)

image registration which enables more precise and quantitative

comparison of retinal disease has not been well developed.

3D SD-OCT imaging technique is a noninvasive and

non-contact scan of retina, which has been widely used in

investigation of retinal pathology, such as Choroidal

Neovascularization (CNV), age-related macular degeneration,

glaucoma and so on [2-6]. SD-OCT acquires information

through one dimension profile scans called A-scans. The

A-scan is constructed by evaluating frequency spectrum of the

interference signal between the reflected light from the tissue

and a reference mirror. SD-OCT device acquires a single

A-scan at a time. By acquiring a series of A-scans in a raster

scanning pattern, the cross sectional slicer B-scan is generated.

Composing successive B-scans yields a 3D SD-OCT image of

retina. Fig.1 shows a 3D SD-OCT scan of retina which consists

of B-scan slices. Each B-scan slice consists of A-scans. Since

SD-OCT imaging is relatively new compared with other

medical imaging modalities such as magnetic resonance

imaging and computed tomography, the requirement for

processing SD-OCT images has a shorter history. With the fast

development of SD-OCT technique, the demand for advanced

image analysis techniques is rapidly growing. Nevertheless, the

development of such techniques can be challenging as SD-OCT

image is inherently noisy and the structure of retina can change

drastically when pathology occurs. Therefore, although many

image registration techniques have been well developed, they

cannot be directly applied to SD-OCT images. The main

reasons are as follows.

1) In SD-OCT, voxels on an A-scan are highly correlated.

For example, the presence of blood vessel in an A-scan will

totally change the gray value of voxels below it. Furthermore,

parallel A-scans in the OCT image actually should be fanning

outwards from the center in physical space which depends on

OCTRexpert：A Feature-based 3D Registration

Method for Retinal OCT Images

Lingjiao Pan,

Fei Shi,

Dehui Xiang

, Kai Yu, Luwen Duan, Jian Zheng and Xinjian Chen

the optics of individual's eye [7]. The geometry of retinal OCT

image makes the deformation across different A-scans

challenging.

2) Eye movements during the scan process lead to significant

spatial distortions (large gaps) between B-scans. These

distortions make the application of existing registration

algorithms difficult especially for interpolation and

regularization.

3) Serious speckle noise of OCT results in poor signal to

noise ratio (SNR) compared with other medical imaging

modalities. The normal intensity based registration methods

tend to be more sensitive to the noises during the registration

process.

4) The layer structures of retina change dramatically in

serious pathological regions. For example, fovea is often

deformed by neovascularization. This makes it difficult to

extract stable features for registration.

5) High resolution of 3D OCT data causes dramatic increase

of the computational complexity for registration and makes the

clinical application difficult.

Fig. 1. Example of a 3D SD-OCT scan of retina.

This paper proposes a new 3D registration algorithm which

can be applied to the longitudinal retinal OCT images for both

normal data and pathological data. Compared with other

existing algorithms, the proposed method presents several

contributions.

1) To the best of our knowledge, this is the first longitudinal

retinal OCT image deformable registration algorithm

which considered both normal data and serious

pathological data.

2) It is a feature-based registration method. In this method,

in order to depress the serious speckle noise in OCT

images, intensity-based region feature, surface-based

structure feature and vessel-like feature are designed to

distinguish different structures of retina.

3) Since the computation complexity of non-rigid registration

is usually very high, a hierarchical deformation mechanism

is designed to reduce local minima as well as to speed up

the computation speed.

II. RELATED WORK

To develop a special registration method for retinal OCT,

many works have been done. Most of the works are restricted to

rigid registration [8-12]. However, it is not enough to describe

the deformation of retina using such a low dimensional model.

Therefore, a deformable registration method is quite necessary.

To the best of our knowledge, limited deformable registration

approaches have been reported for OCT images recently. The

highly-rated deformable registration algorithms such as SyN

and DRAMMS are showed to be unreliable when applied to the

whole OCT image and they took long time (around 8-10 hours)

to finish registration process [13]. Zheng et al. used the SyN

registration algorithm to aid the OCT retinal layer segmentation

[14]. Zhang et al. proposed a two-step image registration

method to reduce speckle noise in OCT images [15]. However,

these registration methods were applied to individual layers or

successive B-scans instead of the whole OCT image. They are

not a full 3D registration. Chen et al. proposed a 3D deformable

registration method using A-scan similarity [13]. They

considered the physical geometry of retina and assumed that the

deformable transformation only occurred along A-scan

direction. In their method, the position of the foveae which was

approximated as the superior point of the thinnest portion of the

retina was used for the initial registration. However, this

method does not work for the data with severe lesions such as

CNV where the fovea is no longer the thinnest portion of the

retina. In our previous work, a two-stage registration method

for OCT volumes with CNV was reported [16]. We used the

projection of 2D vessel points as landmark points and applied

the coherent point drift method followed by the B-spline-based

registration method to find the optimal deformation. This work

was further extended by Du and Gong et al. by using

conditional correlation ratio instead of mutual information as

the similarity measurement for the B-spline-based registration

to improve the registration accuracy [17, 18]. However, these

two methods still have some obstacles. First, their deformable

transformation depends on the intensity similarity and is very

sensitive to the speckle noise especially for the CNV case when

the intensity contrast is low due to the occurrence of

neovascularization. Second, the computation complexity is

usually very high for such intensity-based methods.

In this paper, we propose a feature-based 3D registration

method which can be used for both normal retinal OCT data

and serious pathological OCT data. Although intensity-based

registration methods are more general, feature-based

registration methods have several advantages compared to

them: 1 ） Feature-based methods have low computational

complexity because they evaluate a matching criterion on a

relatively small number of feature points instead of on every

single voxel in an image. 2 ） They can achieve higher

registration accuracy because well selected features have less

ambiguity than intensity similarity in defining correspondences.

3）They show better robustness since the selected features can

avoid the negative impact of image noise effectively. One of the

typical feature-based medical image registration methods is

HAMMER, which is successfully applied to the registration of

brain MRI images and various other organ images [19, 20].

Although HAMMER achieved big success in brain image

registration, it tends to be unreliable when directly applied to

剩余12页未读，继续阅读

评论收藏

内容反馈

weixin_38626858

粉丝: 3
资源: 898

OCTRexpert：A Feature-based 3D Registration Method for Retinal OC...

最新资源

OCTRexpert：A Feature-based 3D Registration Method for Retinal OC...

A method for registration of 3-D shapes

OCR识图工具专业识别图片

Image Registration

Digital Retinal Images for Vessel Extraction

β-Ⅲ-Tubulin: A Reliable Marker for Retinal Ganglion Cell Labeling in Experimental Models of Glaucoma

Shubham-IVth-Year-Fuzzy.zip_REGISTRATION_RETINAL FUZZY_fuzzy seg

三维中绘制feature3D

image registration

Blood-Vessel-Detection-in-Retinal-Images:该项目对眼科使用的各种图像处理技术提供了深刻见解，尤其是使用MATLAB检测视网膜图像中的血管

retinal-exudates-detection:使用混合方法进行渗出液检测（图像形态和机器学习）

SERR-U-Net-Retinal-Vessel-Segmentation:SERR-U-Net

lwnet:最先进的视网膜血管分割与简约模型

matlab视网膜图像分割代码-Luminosity-and-contrast-normalization-in-retinal-images

AE许多插件作用详细解释

ORIGA医疗图像数据集 青光眼

针对视网膜眼底图像中光感受器的高密度小目标检测数据集_A_high-density_sma

A novel method for retinal vascular enhancement

retinal_oct_classification

用卷积滤波器matlab代码-Retinal-OCT-state-of-art:通过OCT重新收集大多数最新的视网膜图像分析论文

深度学习应用于眼科糖尿病视网膜病变预测【原论文】

DRIVE Digital Retinal Images for Vessel Extraction DRIVE数字视网膜图像进行血管提取-数据集

Automatic segmentation of nine retinal layer boundaries in OCT images of non-exudative AMD patients using deep learning and graph search

Retinal Image Analysis

Zebrafish: A tool for behavioral and genetic studies of vertebrate retinal function and degeneration

最新资源

ORIGA医疗图像数据集青光眼