Development of computer-aided diagnosis system for autopsy imaging

首先，我们建立了一个接口系统，将千叶大学CT扫描仪的数据传输到个人电脑进行图像处理，并建立了本研究的图像数据库。鉴于骨折是最重要的死亡原因之一，我们针对头部和胸部的CT图像开发了一套骨折检测系统，主要成果如下：该算法在三维CT图像中检测头部的对称平面，并将左侧的灰度值与右侧相应的灰度值相减，增强差值。利用特征图像法对差值的变化进行统计建模，提出并评价了基于分析结果的骨折检测方法。我们开发了胸部CT图像中骨折检测的流程：a)骨面积提取；b)基于人体解剖的骨分类；c)骨折检测；在此过程中，我们不仅提出了基于二值化的方法，而且提出了基于精细纹理的方法。我们将该方法应用于数据库中的CT图像，并证实它可以成功地提取骨骼区域。过程b)具有点集的统计模型，可以将骨骼划分为解剖单元。从实验结果来看，解剖标记的成功率约为70%。最后一个过程是基于骨表面的连续性检测骨折的位置。实验结果表明，该方法可以检测出70%左右的骨折，并有一些假阳性。

First, we established an interface system for transferring the data from the CT scanner in Chiba University to personal computer for the image processing and built an image database for this study. Then we focused on head and thorax in CT images to develop an bone fracture detection system since bone fractures are one of the most important cause of death Followings are main achievements of this study.1. HeadThe proposed algorithm detected a symmetric plane of the head in 3D CT image and subtracted the gray values in the left part and their corresponding values in the right parts to enhance the difference. The variation of the subtracted values were model statistically using eigen-image method We proposed and evaluated the bone fracture detection method which is based on the results of the analysis.2. ThoraxWe developed the following procedure for detecting bone fracture in the thorax CT imagesa) extraction of bone areab) classification of bones based on human anatomyc) detection of bone fractureIn the process a), we proposed not only the binarization based method but also fine texture based method. We have applied the process to the CT images in the database and confirmed that it could extract the bone region successfully. The process b) has the statistical model of the point set which can classify the bones into anatomical elements. From the experimental results, it was found that the success rate of the anatomical labeling was about 70%. Finally the last process detected locations of bone fracture based on the continuity of the bone surface. The experimental results showed that it could detect about 70% of the bone fractures with some false positives.