【作者】 李居朋；

【导师】 陈后金；

【作者基本信息】 北京交通大学 ， 交通信息工程及控制， 2009， 博士

【摘要】 医学图像处理与分析一直都是图像处理与分析领域中研究的重点和难点问题。由于眼底图像具有特征复杂的特点,使得前人在眼底图像处理与量化分析这方面的研究取得的成果较少,仍存在着一些关键的技术难点,包括多幅不同视野下的图像快速拼接与配准,复杂背景下目标(如视神经乳头、视网膜血管等)精确定位与提取,以及量化参数测量分析等,该方面研究具有较高的科研与临床应用价值。眼底图像的计算机分析系统的建立,可以对眼底组织进行定量测量,在正常和异常之间做出明确鉴别,使眼底系统形态的研究标准化,从而大大加强眼底检查优势。所以本文在此背景下,经查阅大量资料和求教于医学专家基础上,结合临床实际对眼底图像处理及其参数测量方法进行了深入研究,取得的成果总结如下:1.针对眼底图像预处理提出了行之有效的处理方法:提出了基于一次多项式分片逼近几何畸变校正算法和基于彩色信息的辐射量畸变校正算法,解决了山于成像等因素造成的眼底图像畸变;提出了基于mSIFT的眼底图像配准算法,不仅可以实现健康眼底图像的配准与拼接,对病变图像依然有效;同时在眼底图像视场提取等方面做出了研究。2.为准确提取眼底视网膜血管,提出了基于先验知识随机游走模型的视网膜血管分割方法和基于Post-Processing的视网膜血管边缘提取方法,可以很好地实现眼底图像血管分割以及血管边缘提取,特别是针对带有病变的眼底图像,处理效果对比现有算法性能提升明显。3.为描述眼底视网膜血管网络空间属性,提出了一种新的概念——交叉网络,并给出了交叉网络属性测度。充分利用眼底组织间的位置结构关系,提出了基于“类血管”交叉网络的眼底视神经乳头自动定位,多种图像质量下定位成功率测试实验结果验证了定位算法的有效性,运算速度较已有算法也有明显提高。4.视神经乳头形状、面积和深度等参数是衡量眼底健康状况的重要指标,为实现视神经乳头的精确分割和边缘识别,本文充分利用视神经乳头形态特征与色彩特征,提出了基于随机游走模型的视神经乳头分割方法和基于L~*a~*b~*色彩空间视神经乳头边缘提取算法,将图像转换至不同的坐标空间和色彩空间,结合自动定位实现视神经乳头分割与边缘提取。5.为实现眼底图像参数测量,提出了基于三点标尺的计算机图像测量校准与归一化技术,以及长度、面积以及角度等参数测量计算方法。基于对眼底图像处理算法的研究基础上,提出了眼底图像处理与分析系统实现方案,并在WindowsXP平台上,利用Visual C++6.0开发了一套应用系统,该系统实现眼底图像处理、参数测量、辅助诊断以及病历报告输出等功能。对比其它软件,本系统具有更多的图像处理功能,具有较高的实用和推广价值。该研究属于信息学与医学的交叉学科,研究成果很好地解决了眼底图像处理中存在的关键问题,实现了定性和定量分析,提高了临床诊断精度与效率。研制系统可以满足临床眼科工作者的迫切需要,提高我国眼科学诊治水平和人民的健康水平,具有良好的临床应用前景以及社会效益。更多还原

【Abstract】 Medical image processing and analysis plays an important role in image processing and analysis and is generally considered as a difficult problem.Because of the complexity of fundus images,the predecessors as to the quantitative analysis of fundus images got few of the achievements.The fundus images processing and the measurement for its parameters are important to the clinical application and scientific research.The construction of computer aided analysis system of fundus images can be used of quantity measurement of the tissues in fundus images and accurate discrimination between the normal and abnormal images.It can also lead to standardize on the morphological investigation of the fundus system.Also,large advantages have been made on the inspection of the fundus images.In this paper,the fundus images processing technique and the method of its parameters quantitative measurement are studied and many achievements are got.The main achievements are shown as following:1.To improve the quality of fundus images,several pre-processing methods are proposed.Fast distortion rectification method base on slicing approximation and radiant distortion calibration technique base on color information can effectively solve the geometrical and radiant distortion caused by optical lens imaging defects.A robust feature-based method for register and mosaic of the curved color fundus images works well with the rejection error in 0.2 pixels,even for these cases where the fundus images without enough discernble structures.Also,view-field detection for the fundus images is studied.2.To segment the vasculature in fundus images automatically and accurately,a blood vessel segmentation method based on prior knowledge random walks model and an edge detection method based on post-processing are proposed.Experiments indicated that methods perform better than the existing algorithms.Moreover,the efficiency was validated in these results,and could satisfy the requirement of clinic diagnosis.3.To describe the space properties of the retinal vessel,a new conception-cross-network and measures of the network are proposed.Based on the model of fundus organs structure,automatic localization of optic nerve head is realized using the cross-network’s measure parameter-cross density.Experimental results verify the effectiveness of the algorithm using the fundus image databases and the clinic images over different image qualities.And it can satisfy the requirement of clinic ophthalmological diagnosis.4.The shape,area and depth of the optic nerve head are important measurement indexes of Health Status for fundus.To realize the computer-aided measurement of these characters,an automated method to segment the optic nerve head based on the improved Random walks segmentation method and an automated detection of optic nerve boundary based on L~*a~*b~* color space are proposed in this paper.In contrast with the state-of-art algorithms,a better overlap accuracy is acquired in the comparing segmentation results for our clinic fundus images database with the golden standards and verifies the effectiveness of the proposed algorithm.5.To realize images calibration and parameters measurement,a novel calibration method based on three points scale and parameters measurement method for length, area and angle are proposed.On the base of the research of fundus images processing algorithm,a kind of developing scheme on fundus image processing system is presented. A software is developed using Visual C++6.0 under the Microsoft windows XP operation system.The software can realize many functions such as fundus image parameters quantitative measurement,assistant diagnosis,the geometry transform of fundus images,gray transform,image enhancement and the pathology report output.In contrast to other software,this software has more functions of image processing and realizes the measurement of many parameters with high application value.The research project of this paper is an intersectant research area,which is based on the development of rapid information discipline and biomedical technology.Research results solved the key problems in the fundus images processing,realized qualitative and quantitative analyses and improved precision and efficiency of clinical diagnosis. The developed system has good prospective clinic application and social benefit,as it meet the requirement of ophthalmologists,enhance ophthalmology diagnosis treatment levels and improve people health status.更多还原