Takahashi K, Yonezawa H, Chiba K, Tohgi H
Medical Computer Research Center, School of Medicine, Iwate Medical University, Honcho-dori 3-16-1, Morioka 020, Japan.
Medinfo. 1995;8 Pt 2:911.
Statistically identified information on the relationships between the sites of lesions in intracerebral hemorrhage (ICH), risk factors such as a smoking or drinking habit, anamnesis, and biochemical data through blood tests will extend assistance to neuromedical clinicians on their daily clinical duties. It will provide them with a useful guide to determine the method of treatment. Also, it will be a basic research material for their clinical studies on diagnosis, progress, or prognosis in ICH. In order to obtain such statistics with the help of the computer, we need to have a computationally effective image database system. As is generally known, medical image data especially requires a great amount of storage; high-speed processing techniques are therefore also needed to deal with such data effectively. In addition, it is desired that we have outputs from the analysis edited with well-visualized effect, using 3D computer graphics, etc. These are why most existing image processing systems have been designed to work on comparatively large-scale computers. So far as we know, it is hard to find a practical and inexpensive personal computer-based application system for visualized statistical analysis of lesional images in ICH. We have developed a desk top computer-based program for statistical analysis of lesional image data of ICH. With this system, we can organize a medical image database that consists of the personal data of patients with ICH (sex, age, occupation, diagnosis, symptoms, part of physical disorder, etc.), risk factors, anamnesis (cerebral apoplexy, hypertension, hypotension, corpulence, diabetes, hyperlipidemia, atrial fibrillation, valvular endocarditis, etc.), biochemical data of blood, and lesional image data from CT or MRI. This system consists of the following components: 1) database management, 2) information retrieval (IR), 3) lesional image processing, 4) statistical analysis, and 5) prognostic prediction. The images are drawn manually on prescribed data sheets by tracing CT or MRI films and are read through the image scanner; then the compressed data of the digitized images is recorded in the database. Each recorded image data consists of the following two components: the frame image that corresponds to the contour of tissues of interest on the corresponding sliced section, and the actual image that corresponds to the lesion itself. In our system, these two images are separately stored and managed so that we can effectively perform subsequent image analysis. Other variables in the database (risk factors, anamnesis, etc.) are mainly used as search keys for making the aggregate of image data by the IR subsystem. In any aggregate, its elements, namely image data, have common medical background descriptions with the search keys. These aggregates can be used as input for the lesional image processing subsystem. With this subsystem, we can obtain the accumulated distribution of frequencies within a specified range of any sliced section, display planar color maps and profiles associated with the distribution, reconstruct it in 3D form, perform transformations of 3D images (zooming, enhancement, rotation, etc.), and test the significant difference of frequencies between any two different sites. We have been making practical use of this system to find the neurological relationship between the symptom (dysarthria, and paralysis of upper/lower limbs) and the site of lesion with cerebral infarction in pons. This study is quite important since the distributions of pyramidal tract related to the above symptom in pons are not well-known compared to those in cerebral cortex, internal capsule, or cerebral peduncle. With our system, we have obtained several findings expected to be helpful for this study. However, since this study is still in the initial phases, we will only present the outcome as a working example of our system. Our system was originally developed for analyzing lesional images with ICH. However, it could
通过统计确定脑出血（ICH）病变部位之间的关系、吸烟或饮酒习惯等风险因素、既往史以及血液检测的生化数据，将有助于神经医学临床医生的日常临床工作。这将为他们提供确定治疗方法的有用指南。此外，它还将成为他们关于ICH诊断、进展或预后临床研究的基础研究材料。为了借助计算机获得此类统计数据，我们需要一个计算高效的图像数据库系统。众所周知，医学图像数据尤其需要大量存储空间；因此还需要高速处理技术来有效处理此类数据。此外，希望使用3D计算机图形等对分析结果进行编辑，使其具有良好的可视化效果。这就是为什么大多数现有的图像处理系统都设计在规模相对较大的计算机上运行。据我们所知，很难找到一个实用且廉价的基于个人计算机的系统来对ICH病变图像进行可视化统计分析。我们开发了一个基于台式计算机的程序，用于对ICH病变图像数据进行统计分析。通过这个系统，我们可以构建一个医学图像数据库，该数据库由ICH患者的个人数据（性别、年龄、职业、诊断、症状、身体疾病部位等）、风险因素、既往史（脑中风、高血压、低血压、肥胖、糖尿病、高脂血症、心房颤动、瓣膜性心内膜炎等）、血液生化数据以及CT或MRI的病变图像数据组成。该系统由以下组件组成：1）数据库管理，2）信息检索（IR），3）病变图像处理，4）统计分析，5）预后预测。通过追踪CT或MRI胶片，在规定的数据表上手动绘制图像，然后通过图像扫描仪读取；接着将数字化图像的压缩数据记录在数据库中。每个记录的图像数据由以下两个组件组成：与相应切片上感兴趣组织轮廓相对应的框架图像，以及与病变本身相对应的实际图像。在我们的系统中，这两个图像分别存储和管理，以便我们能够有效地进行后续图像分析。数据库中的其他变量（风险因素、既往史等）主要用作搜索键，以便IR子系统对图像数据进行汇总。在任何汇总中，其元素，即图像数据，与搜索键具有共同的医学背景描述。这些汇总可以用作病变图像处理子系统的输入。通过这个子系统，我们可以获得任何切片在指定范围内频率的累积分布，显示与该分布相关的平面彩色图和剖面图，以3D形式重建它，对3D图像进行变换（缩放、增强、旋转等），并测试任意两个不同部位之间频率的显著差异。我们一直在实际使用这个系统来寻找桥脑脑梗死症状（构音障碍以及上肢/下肢麻痹）与病变部位之间的神经学关系。这项研究非常重要，因为与大脑皮层、内囊或脑桥相比，桥脑中与上述症状相关的锥体束分布尚不清楚。通过我们的系统，我们已经获得了一些预期对这项研究有帮助的发现。然而，由于这项研究仍处于初始阶段，我们将仅把结果作为我们系统的一个工作示例呈现出来。我们的系统最初是为分析ICH病变图像而开发的。然而，它可以
