【作者】 魏平；

【导师】 鲁东明； 陈刚；

【作者基本信息】 浙江大学 ， 计算机科学与技术， 2009， 博士

【摘要】 乳腺钼靶图像是诊断乳腺癌的重要影像手段,而数字乳腺钼靶图像的软拷贝阅片方式越来越广泛地取代了传统的胶片硬拷贝阅片方式,数字乳腺钼靶图像的软拷贝显示技术对诊断质量有重要影响,因此倍受关注,成为医学图像处理的重要的研究领域之一。本文首先介绍了数字乳腺钼靶图像软拷贝显示技术的分类,研究现状及性能分析。目前采用的软拷贝显示技术没有紧密结合乳腺钼靶图像的特征和诊断先验知识,不足以应对乳腺钼靶图像的低对比度和高分辨率给图像显示带来的挑战。现有方法的不足主要表现在:首先,乳腺区域分割技术不能满足软拷贝显示预处理同时要求的高精度和高速度;其次,在对乳腺钼靶图像进行锐化增强时,过冲现象和噪声敏感性始终难以避免,没有突显出对重要的诊断信息的增强效果;再次,窗密度显示技术存在初始显示对比度、亮度不高的问题,而采用直方图规定化方法又会产生难以接受的信息损失,或造成窗密度交互显示的不连续;最后,窗密度交互显示最耗时的计算部分没有得到优化,实时性能仍然有很大的提升空间。为了解决以上问题,本文就数字乳腺钼靶图像的软拷贝显示技术进行了深入的研究。主要研究内容如下:1.基于等值面特征的乳腺区域分割算法乳腺钼靶图像的等值面特征是由成像机制决定的,具有相当的稳定性。以等值面特征为依据计算分割阈值,通过基于线段的区域生长获得初步的乳腺区域后,又对未曝光边框进行了剥离,最终提取到了比较纯粹的乳腺区域。每个步骤的计算都具有较低的计算复杂度,执行效率很高,且提取的乳腺区域具有较高的精度。2.基于模糊逻辑的多尺度乳腺钼靶图像锐化增强算法为了克服增强算法对噪声敏感的缺点,设计了基于模糊逻辑的混合滤波器对脉冲噪声和高斯噪声进行抑制,考虑到乳腺钼靶图像的诊断信息蕴含在不同尺度上,采用Laplace金字塔对图像进行多层分解。计算子带图像增益系数时,体现了局部对比度及乳腺组织的不同重要性。算法克服了易出现的过冲现象和对噪声敏感性,从视觉质量上和客观评价指标上都具有明显优势,对乳腺钼靶图像的诊断信息有很好的增强效果。3.模拟动能变化过程的窗口参数优化算法窗口参数的选择对医生的阅片质量有重要影响。将乳腺区域像素值概率密度直方图视作崎岖不平的地形,优化窗口参数的计算过程模拟为小球运动时的动量在上坡、下坡、受阻到停止的一系列变化过程,动量的最大点对应优化的窗口参数。该算法具有很好的鲁棒性,能明显改善乳腺钼靶图像初始显示时乳腺区域的显示亮度。4.基于动态范围转换的窗密度显示算法给出了动态范围转换函数的定义并对相关理论进行了论证,以此为基础构建了动态范围转换算法。动态范围转换函数体现了像素值出现频率及不同组织密度在诊断中的重要性,是一种像素值到可显示灰度的非线性转换,能够有效改善图像对比度。不存在信患损失,也能保持窗密度交互时的显示连续性。5.窗密度交互显示的实时优化算法该方法利用乳腺区域的分割结果在调窗时避免对背景区域像素的计算。从理论上证明了二分法逐步逼近思想进行插值优化的可行性,据此给出了图像双线性快速插值算法,将耗时的浮点运算用整数运算和位运算代替,并去除了冗余运算。快速插值算法的计算效率提高了10倍左右,乳腺钼靶图像软拷贝显示窗口调节的实时性得到显著提高。上述算法都通过实验验证了有效性,大部分成果已经集成在PACS系统中,目前有数十家医院使用,取得了很好的效果。更多还原

【Abstract】 Mammogram is an important imaging tool to diagnose breast cancer. Film mammogram based hard copy reading is widely replaced by digital mammogram based soft copy reading. Soft copy display has great infulence on diagnosis, so it achieves wide attention and becomes one of the most important research areas of medical image processing.In the first part of this dissertation, we introduce the classification, present research situation and performance analysis of digital mammography soft-copy display techniques. The present used soft copy display techniques didn’t fully use priori knowledge of diagnosis and characters of digital mammograms, so these techniches are not good engough to confront the challenges of image presentation arrised by high resolution and low contrast of digital mammograms. The main shortcomings of present techniques include: Firstly, the breast region segementation techniques didn’t satisfy the demand of high accuracy and high speed for soft copy display preprocessing. Secondly, the unsharp enhancement methods have disadvantages of over shooting and noise sensitivity, import information for diagnosis is not effectively enhanced. Thirdly, window intensity techniques have problems of low contrast and low luminosity, while histogram specification methods will either introduce unbearable information loss or cause discontinuity presentation during interactive display. At last, the most time consuming part of window intensity interactive is not optimized, so still has great space to improve the real-time capability.In order to solve those problems mentioned above, this dissertation focuses on digital mammogram soft-copy display. The main research topics are as follows:1. Thresholding method based on identified intensity planesThe features of identified intensity planes in mammogram are dominated by mammography mechanism, they are very stable and been used to calculate the threshold to segment breast region and background. After obtaining rough breast region by line segment based region growing method, a refine procedure is used to peel off the unexposed band regions. At last, pure breast region is extracted. Precision and efficiency of the method have been verified by experimental results.2. Fuzzy logic based mutiscale mammogram enhancement methodIn order to strengthen noise resistance, the original image is filtered by the hybrid low pass filter to remove out the impulse noise and Gaussian noise. Considering diagnosis information contained in different scales, the filtered image is decomposed by Laplacian pyramid. During enhancement coefficient calculation, local contrast and importance of different tissue are considered. The proposed methed overcomes over shooting and noise sensitivity, it can achieve good performance for mammogram enhancement and shows obvious advantage at both visual quality and objective indicator.3. Window parameter optimization by simulating the process of kinetic energy variationWindow parameter setting has great influence on mammogram reading. In this method, the probality intensity histogram of breast region is treated as a bumpy road and the procedures of calculating the optimal window parameters are simulated as the process of kinetic energy variation, just like a small ball rolling through the bumpy road, including up/down slope, resistance and finally stoped. The window parameters can be calculated when the kinetic energy of the small ball achieves the maximum value. The proposed mehod has good robustness; it can improve intial display luminance of breast region in mammogram.4. Window intensity display method based on dynamic range transformationWe improved window intensity display by introducing the concept of dynamic range transformation (DRT). The formal definition of DRT is presented and relative theorems are proved. DRT is a nonlinear transformation from pixel intensities to gray levels which reflects the frequency of occurences of pixel intensities and importance of different breast tissues. By using DRT, the mammogram display quality can be effectively improved. In addition, DRT doesn’t cause information loss and keeps display continuity at interactive window intensity operations.5. Real-time window intensity interactive display optimizationTo avoiding calculation of pixels in background, the method uses breast region segmentation result. The feasibility of optimizing polynomial interpolation with dichotomy is proved. A fast bilinear interpolation method for image zooming is prosed. The method replaces floating-point operation by integer operation and bit-shift operators. Compared with the original method, the excuting speed increases almost 10 times and the real-time capability is obviously improved.All methods metioned above have been verified by experiments. Most of them have been intergrated into PACS which were used in dozens of hospitals and have achieved good benefits.更多还原