【作者】 郝世杰；

【导师】 蒋建国； 詹曙；

【作者基本信息】 合肥工业大学 ， 信号与信息处理， 2009， 硕士

【摘要】 健康是现代人类社会主题之一。作为人类智慧和科技的结晶,医学影像学使得我们可以非介入式地观察人体内部构造和诊断、治疗疾病,从而使得人类医学事业向前迈进了决定性的一步。而医学图像处理与分析作为一项以信息科学技术为基础的交叉学科研究,将注定使医学向数字化、智能化、自动化方向再迈进一大步。本文针对脊椎核磁共振图像,开展图像分割处理、特征定量提取分析和脊椎医学图像三维可视化的研究。主要工作和创新点如下:(1)提出改进型主动外观模型分割脊椎核磁共振图像中椎体结构。经典主动外观模型建立中假定训练数据服从高斯分布前提,导致使用主分量分析生成的模型对物体边缘和纹理细节描述不足的局限性,用独立分量分析代替主分量分析,分别建立形状模型和纹理模型,进而建立细节描述能力更佳的改进型主动外观模型,使用此模型分割核磁共振图像中的腰椎椎体。(2)在腰椎椎体分割的基础上,对图像里反映的脊椎生理和病理特征提出一些定量提取与分析算法。首先是基于最小描述长度准则确定表现腰椎弯曲曲线的最佳拟合函数,进而提出具有旋转、平移、缩放不变性的曲率算子计算椎体间的腰椎弯曲成角,代替传统通过椎体终板延长线交角测量腰椎弯曲的方法;其次通过确定椎间盘局部区域的倾斜灰度投影确定椎间盘退行性变化程度,并以此作为马尔科夫随机场分割椎间盘的区域分类依据,实现快速无监督椎间盘分割;最后结合腰椎弯曲曲线,定量计算腰椎间盘突出程度。(3)在可视化算法平台VTK基础上,使用游走立方体算法和光线跟踪算法分别对脊椎计算机断层图像和核磁共振图像进行面绘制和体绘制的三维可视化。各部分实验结果证明,改进型主动外观模型分割精度有一定提高,能够确保后续工作的准确性;基于最佳拟合曲线和曲率算子定量计算出的腰椎弯曲成角更为客观和准确;可在保持椎间盘分割精度前提下大大提高分割速度;计算出的椎间盘突出程度对病变部分给出了量化描述,对临床诊断和观察具有参考价值。更多还原

【Abstract】 Health is one of the main topics in our modern society. As the production of technology and human wisdom, medical imaging, which provides a non-intervention overview of inner structures of our bodies and great help in diagnosis and cure, has pushed medical society significantly. Combining fields like information science and medicine, medical image process and analysis is making another push in the progress of medicine digitally, automatically and intelligently.In this paper, research is done on segmenting, quantitative analyzing and visualizing spine MRI. The main points are as follow:(1) It is proposed that the vertebral structure is segmented with improved active appearance model (AAM). Considering the assumption of Gaussian distribution of training data in classic AAM, the model built with principle component analysis (PCA) is insufficient in describing the details of shape and texture. So PCA is replaced by independent component analysis (ICA) in building shape model and texture model in AAM model building. And the spinal vertebral bodies are segmented with this improved AAM.(2) Based on segmenting spinal vertebral structure, a series of quantitative extraction and analysis are presented, aiming at obtaining quantitative physiological and pathological features in spine MRI. Firstly, lumbar curve is optimally fitted based on minimum description length (MDL); curvature operator is then introduced into computing lumbar lordosis angles. Secondly, the distribution of intensity in local lumbar disc region is detected through intensity projection; based on this information, lumbar discs are segmented with a fast and unsupervised Markov random field (MRF). Thirdly, the quantitative lumbar disc herniation is evaluated with the cooperation of lumbar curve and segmented discs.(3) On the visualization platform provided with Visualization ToolKit (VTK), spine CT and MRI are three-dimensionally visualized respectively with marching cubes and ray casting algorithm.The experiments of each point are made. The segmentation error with improved AAM is lower than that of classic AAM. The optimal lumbar curve and lordosis angles are subjective and accurate. In segmenting lumbar disc, the temporal cost is much lower while segmentation accuracy is kept. The quantitative lumbar disc herniation is of great clinical value.更多还原