Connection and Deformation of Pathological Images via a Macro Image for Comparing Different Modality Images of Brain Tumor
Connection and Deformation of Pathological Images via a Macro Image for Comparing Different Modality Images of Brain Tumor
复制标题
DOI：
10.1155/2014/368951
复制
发表时间：
2014-11-19
期刊：
影响因子：
--
通讯作者：
Yagi Y
中科院分区：
文献类型：
--
作者：
Ohnishi T;Tanaka T;Nakamura Y;Hashimoto N;Haneishi H;Taylor J;Snuderl M;Yagi Y
背景磁共振成像(MRI)是诊断脑肿瘤的首选检查方法。然而，由于肿瘤的浸润区在MR图像上往往是模糊的，因此很难准确地识别肿瘤区域。为了揭示肿瘤的组织信息与MR信号之间的关系，必须对同一区域的病理图像和MR图像进行分析。然而，由于病理图像是通过制作组织标本来分割和变形的，所以将病理图像与磁共振图像联系起来并不容易。提出了一种参考光学相机拍摄的宏观图像对一组病理图像和MR图像进行配准的方案。第一步，将分割后的病理图像与宏观图像进行拼接和变形。第二步，将已连接的病理图像与MR图像配准。阐述了病理图像与宏观图像的连接和变形方法。图1显示了建议方法的流程。在连接步骤之前，将每一幅图像转换为含有病理图像的绿色分量或宏观图像的蓝色分量的单色，然后人工对病理图像的初始位置进行粗略校正。然后，利用对应的特征点对同一切片上的所有病理图像进行连接。这里，特征点是从要连接的边缘区域中手动选择的。在连接特征点的同时，用薄板样条法对其他区域进行变形。之后，参考宏观图像对连接的病理图像进行变形。变形步长包括基于地标的阶段和基于强度的阶段。在基于标志点的匹配阶段，从病理图像和宏观图像中选择标志点，并以类似于连接步骤的方式对它们进行匹配。最后，采用基于灰度的匹配方法对病理图像进行变形。用归一化互信息评价两幅图像的相似性，并用Powell-Brent方法最大化。
BackgroundMagnetic resonance imaging (MRI) is a preferred modality for diagnosis of brain tumor. However, because infiltrated regions with tumor are often indistinct on the MR image, it is difficult to identify tumor regions exactly. For revealing the relationship between tissue information and MR signal of the tumor, pathological images and MR images at the same regions have to be analyzed. However, it is not easy to make relationship between pathological images and MR images because pathological images are divided and deformed through tissue specimen making. We propose a registration scheme of a set of pathological images and MR image by referring a macro image captured by an optical camera. In the first step, parted pathological images are pieced together and deformed with macro image. In the second step, connected pathological image is registered to MR image. This paper shows connection and deformation methods for the pathological image.MethodThis paper explains a method for the connection and deformation method of the pathological image with the macro image. Figure 1 shows the flow of the proposed method. Before the connection step, each image is converted to monochrome with green component for the pathological image or blue component for the macro image, and then initial positions of pathological images are roughly corrected manually. Next, all pathological images at the same slice are connected by use of corresponding feature points. Here, feature points are manually selected from edge region to be connected. While feature points are connected, other regions are deformed by thin plate spline technique. After that, the connected pathological image is deformed referring to the macro image. Deformation step consists of landmark based and intensity based phases. In the landmark based phase, landmarks are chosen from the pathological image and macro image and match them in a similar way to the connection step. Finally, pathological image is deformed using intensity based matching method. Similarity between both images was evaluated by normalized mutual information and maximized by Powell-Brent method.