【作者】 李前娜；

【导师】 孙丰荣；

【作者基本信息】 山东大学 ， 通信与信息系统， 2007， 硕士

【摘要】 冠心病已成为目前人类疾病死亡的主要原因之一。90年代末期诞生的冠心病诊断新技术—实时心肌声学造影(Real Time Myocardial ContrastEchocardiography,RT-MCE)作为一种能够无创和定量评估心肌微循环的新技术,正在从基础研究走向临床实践,如何提高MCE图像定量分析技术的准确性和客观性已成为当前国内外相关学科领域的研究热点。但是,目前所有的心血管超声设备或软件都没有解决基于MCE冠心病定量分析的两个关键技术问题:MCE心脏超声序列图像的弹性配准和MCE图像的降噪。这两个问题影响了MCE技术对冠心病诊断的准确性和客观性,且阻碍了MCE作为一种冠心病无创诊断新技术临床地位的确立。高质量的图像是成功进行MCE冠心病定量分析的前提和保证,但MCE图像存在严重的Speckle噪声,它有着非常复杂的形成机理和统计规律。在以往的很多医学超声图像降噪方法中,探讨MCE图像降噪的文献较为少见,且国内尚未有研究见报道。已有的MCE图像降噪方法包括非线性小波变换滤波方法,非线性小波扩散滤波方法,有偏的二维最小均方滤波方法等。上述研究都有着很强的理论探索意义和学术价值,但临床应用效果都不是非常理想,因此,需要探讨研究新的理论方法来解决这一问题。加拿大学者R.G Stockwell博士于1996年提出了S-变换(S-Transform)理论,并且完备地证明某函数的S-变换可以由其小波变换经相位校正得到,即S-变换在多分辨率框架下可以被看作是小波变换的一种扩展。S-变换分析技术(S-TransformAnalysis)融合了短时Fourier变换与小波变换的优点,有望成为多分辨率分析和非平稳随机信号处理强有力的新工具。小波变换在解决医学图像降噪与增强问题上有着出色的表现,因此S-变换分析技术势必在医学图像降噪与增强领域有着更为巨大的应用前景。基于此,我们参考阅读了大量文献,深入研究了S-变换分析的理论、技术和算法实现,并参考小波变换阈值降噪方法的原理,经过多次探索试验,得出了基于S-变换分析技术的图像降噪与增强算法,进而设计出了性能更加出色的在S-变换分析技术框架下抑制MCE图像Speckle噪声的算法。对于本课题提出的MCE图像Speckle噪声抑制算法,作者采用临床取得的实际MCE图像进行了试验,试验结果表明,和以往的降噪方法相比,本方法更能有效地抑制MCE图像中的Speckle噪声,在MSE,SMSE两种通用评判乘性噪声的定量标准上,基于S-变换的降噪方法都取得了良好的应用效果。同时,由于图像的2-D S-变换结果是个4维的结构,其总体计算复杂度为O(N4logN),与此同时,2-D S-变换所需的内存增长为O(N4)。由此可见,2-D S-变换的运算量和所需内存都较大,例如,对于一个256×256的图像,所需要的内存空间就是2564。这样的内存需求足以扼杀S-变换,就像没有快速算法的Fourier变换一样。由于尚未有2-D S-变换的快速算法见诸报道,因此,本课题决定采用并行计算的途径来解决上述两大问题;经过调研,作者使用了Linux下的微机集群作为并行计算的平台。更多还原

【Abstract】 Coronary Disease has been one of the main death reasons for human being. Real Time Myocardial Contrast Echocardiography (RT-MCE) is a new hurtless technique for quantitative estimate on cardiac microcirculation, and is starting to be introduced into clinical practice. There are two key problems for MCE Coronary Disease quantitative analysis: Register of MCE heart sequence images and the despecking of MCE image. Until now, all the reported quantitative analysis software systems for MCE haven’t solved them.High quality image is the precondition and assurance for MCE Coronary Disease quantitative analysis. MCE image contains serious speckle noise, and it is a kind of very complex multiplicative noise. There are many literatures on despecking of medical ultrasound image, but the one for MCE image is very limited. The methods about despecking of MCE image contain: nonlinear wavelet shrinkage techniques, nonlinear wavelet diffusion method, biased motion-adaptive temporal filtering method. All the researches above have great theory meaning and academic value, but the clinical practice effect doesn’t satisfy very well. So, it need study new method to solve this problem.R. G. Stockwell, a Canadian scholar, proposed the 1-D S-Transform theory in 1996. He has proved that we can derive S-transform as the "phase correction" of the continuous wavelet transform. So, in the frame of multiple resolution, S-Transform is an extension of wavelet transform. As a new local multi-scale spectrum analysis technology, S-Transform has distinct advantages then wavelet transform and short time Fourier transform (STFT). As a result, S-Transform analysis technique must has great potential on medical image despeckling and enhancement. So, we decided to study S-Transform analysis technique and explore its application on despeckling of MCE images.In this case, we have studied the theory, technique, and the realization of S-transform in-depth. At the same time, through considering of wavelet transform, we developed noise reduction and contrast enhancement algorithm based on S-transform analysis technique. Further more, we designed a more effective speckle noise reduction and contrast enhancement algorithm for MCE image based on S-transform analysis technique. For this algorithm, we did experiments on clinical practice MCE images, and the results of MSE, SMSE estimating standards show us that our method does better than other several methods on speckle noise reduction.At the same time, the S-Transform of a 2-D image produces a four-dimensional structure, so, the overall complexity for the 2-D S-Transform is O[N~4 log(N)], and the memory requirement grows as N~4 . Thus, a 256*256 image requires 256~4 storage elements, 32 GB of floating point values for example. These memory requirements pose problems not only for ultimate long-term storage but also for the execution of an ST, like the Fourier Transform without fast algorithm. So, in the absence of fast ST algorithm, we decided using parallel computing to solve these two problems; after investigation, we chose Linux PC cluster as the parallel computing platform.更多还原