计算机辅助桡骨骨龄等级评估

发布时间：2018-05-09 08:43

本文选题：骨龄评估 + 中华05　；参考：《北京交通大学》2017年硕士论文

【摘要】：骨龄是评估青少年儿童体格发育程度的重要指标,在体育科学、司法和临床医学等领域有着广泛的应用。由于手腕骨可以较为准确的反应整体骨骼生长发育情况,并具有易于拍摄、辐射剂量小等特点,因此国内外普遍将手腕骨发育情况作为骨龄评价标准。中华05计分法是目前我国对手腕骨骨龄评估所使用的标准方法。但使用该标准进行评估需要对人员进行专业培训,熟练掌握各个发育分期的特征,整个过程十分繁琐,而且存在主观性较强、精确度低的缺点,因此实现计算机自动评估骨龄的需求不断增长。目前计算机自动骨龄评估主要存在两个困难。首先,CT图像中手腕骨出现的位置、方向和大小不确定,并且在骨龄后期骨块之间会出现重合的现象,对提取骨块的分割与提取造成干扰。其次,骨龄标准中由人类语言所描述的不同等级生长发育特征难以转换成被计算机处理的图像特征。针对以上问题,本文提出了一种基于中华05计分法的计算机辅助桡骨等级评定方法,针对手腕骨CT图像的桡骨特点,对桡骨等级2至7级的自动评估过程进行了研究。本文的主要内容如下:(1)提出了一种基于多模板约束局部模型的桡骨分割方法。针对不同骨龄阶段桡骨形状变化比较剧烈,约束局部模型不易收敛的问题,提出一种多模板的约束局部模型算法,通过对桡骨图像进行预分类并与真实桡骨形状接近的桡骨模型进行分割。通过实验表明,该算法具有更快的收敛速度和更高的分割准确度。(2)针对CT图像背景单一的特点,本文提出一种Haar随机森林算法并将其作为约束局部模型的局部检测器。通过实验表明,相比使用PCA或SVM作为局部检测器的约束局部模型,该算法在分割CT图像中的桡骨区域时具有更高的鲁棒性和准确度。(3)根据中华05计分法的桡骨等级评定标准选取了部分具有代表性的灰度特征和局部形状特征,同时根据手腕骨CT图像中桡骨特点,提出了基于PCA的全局形状特征和基于尺度变换不变纹理特征。然后设计了基于随机森林的桡骨等级分类器,并通过实验对优化模型参数,验证了本文所提取的骨龄特征的分类性能。
[Abstract]:Bone age is an important index to evaluate the physical development of adolescents and children. It is widely used in sports science, justice and clinical medicine. Because the wrist bone can accurately reflect the whole bone growth and development, and has the characteristics of easy shooting and low radiation dose, the development of the hand wrist bone is generally regarded as the bone age evaluation standard at home and abroad. Zhonghua 05 scoring method is the standard method used to evaluate the bone age of the wrist bone in our country. However, the use of this standard for evaluation requires professional training of personnel, proficiency in the characteristics of various stages of development, and the fact that the whole process is cumbersome and has the disadvantages of being highly subjective and of low accuracy. Therefore, the demand for automatic evaluation of bone age by computer is increasing. At present, there are two main difficulties in automatic bone age assessment by computer. Firstly, the position, direction and size of the wrist bone in CT images are uncertain, and there will be overlap between the bone fragments in the later stage of bone age, which interferes with the segmentation and extraction of bone fragments. Secondly, the characteristics of growth and development of different grades described by human language in bone age standard are difficult to be converted into image features processed by computer. In order to solve the above problems, a computer-aided radial grading method based on Zhonghua 05 scoring method is proposed. According to the radial characteristics of CT images of wrist bone, the automatic evaluation process of radial grade 2 to 7 is studied. The main contents of this paper are as follows: (1) A method of radius segmentation based on multi-template constraint local model is proposed. A multi-template constrained local model algorithm is proposed to solve the problem that the radial shape changes dramatically at different bone ages and the constrained local model is not easy to converge. The radial image was preclassified and segmented from the real radial model. Experiments show that the algorithm has faster convergence speed and higher segmentation accuracy. Aiming at the single background of CT images, this paper presents a Haar stochastic forest algorithm and uses it as a local detector for constrained local models. The experimental results show that compared with using PCA or SVM as the constrained local model of local detector, The algorithm has higher robustness and accuracy in segmenting radius area in CT image. According to the radial grade evaluation standard of Zhonghua 05 score method, some representative gray scale features and local shape features are selected. According to the characteristics of radius in wrist CT images, the global shape feature based on PCA and the invariant texture feature based on scale transformation are proposed. Then, a radial classifier based on random forest is designed, and the model parameters are optimized by experiments to verify the classification performance of the bone age features extracted in this paper.
【学位授予单位】：北京交通大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：R68;TP391.7

【相似文献】