基于随机森林算法的小鼠micro-CT影像中骨骼关节特征点定位

发布时间：2018-06-04 22:49

本文选题：小动物影像分析 + 骨关节点定位　；参考：《中国生物医学工程学报》2017年03期

【摘要】：随着小动物成像技术的发展,技术人员每天需要处理的小动物影像数量急剧增长,这使得自动化的小动物图像分析方法成为迫切的需求。在小鼠图像分析方面,小鼠灵活多变的身体姿态给自动化的图像分析带来困难。基于随机森林算法实现小鼠micro-CT图像中骨骼关节点的自动定位,为解决小鼠影像中身体姿态的自动识别打下基础。该算法主要分3步:先通过分类随机森林算法得到小鼠骨骼关节点的粗定位,再通过回归随机森林算法进一步减小定位误差,最后通过图匹配的方法在备选点中挑选正确位置上的关节点。对49例不同身体姿态的小鼠全身三维micro-CT图像进行测试,全身关节点定位的成功率为98.27%,定位误差的中值为0.68 mm。同时验证联合使用分类与回归随机森林的必要性,并探究训练数据的数量对不同骨关节的识别效果的影响。研究为小鼠micro-CT影像中身体姿态的识别提供一种新方法,为后续的自动化图像配准、图像分割以及自动化图像测量提供重要的定位信息。
[Abstract]:With the development of small animal imaging technology, the number of small animal images that technicians need to deal with every day increases rapidly, which makes the automatic small animal image analysis method become an urgent need. In the aspect of image analysis of mice, the flexible and changeable body posture of mice makes it difficult to automate image analysis. Based on the stochastic forest algorithm, the automatic location of the skeletal node in mouse micro-CT image is realized, which lays the foundation for automatic recognition of body posture in mouse image. The algorithm is mainly divided into three steps: firstly, the rough location of mouse skeletal node is obtained by classifying stochastic forest algorithm, and then the localization error is further reduced by regression stochastic forest algorithm. Finally, the correct position of the node is selected in the alternative point by the method of graph matching. Three-dimensional micro-CT images of 49 mice with different body posture were tested. The success rate of locating the whole body knots was 98.27 mm, and the median of positioning error was 0.68 mm. At the same time, the necessity of combined use of classification and regression random forest was verified, and the effect of the amount of training data on the recognition effect of different bone joints was explored. This study provides a new method for the recognition of body posture in mouse micro-CT images, and provides important location information for subsequent automated image registration, image segmentation and automatic image measurement.
【作者单位】：大连理工大学生物医学工程系;
【基金】：国家自然科学基金(61571076);国家自然科学基金青年基金(81401475) 辽宁省自然科学基金(2015020040)
【分类号】：R814;TP391.41

【相似文献】