Research on 3-D analysis of histopathologic specimen for histopathologic diagnosis support

本研究的目的是将组织病理标本作为三维对象进行分析，并开发一个组织病理学诊断支持系统，该系统可以计算出与标本厚度无关的三维核密度等有用信息。从理论和实验上评估了标本厚度对观察到的(2-D)核密度的影响，发现1um的厚度变化会导致约10%的核密度变化。实现了一种非破坏性测量组织病理学标本厚度的方法。这种方法使用光学显微镜捕获的图像，以便许多病理学家使用它。使用光学显微镜观察的组织病理图像通常具有不希望看到的模糊。实现了一种降低三维图像模糊的方法。该方法采用改进的盲解卷积方法，对原子核进行三维建模，获得了满意的结果，实现了一种估计原子核参数(三维原子核密度、原子核半径的平均值和方差)的方法。该方法只利用了病理学家通常观察到的二维显微图像，因此他们可以很容易地使用该方法。为了减少计算时间，提出了逐次估计各参数的序贯估计方法，并开发了图形用户界面，使用户可以方便地使用该方法。该图形用户界面具有半自动提取核位置和核等值线、估计核参数的功能。通过简单的鼠标操作，核的位置和核的轮廓可以很容易地纠正，所以病理学家可以很容易地估计出与标本厚度无关的3-D核密度。

The purpose of this research were to analyze a histopathologic specimen as a 3-D object, and to develop a histopathologic diagnosis support system which calculate useful information such as 3-D nuclear density independent of the specimen's thickness.Effect of specimen's thickness on the observed (2-D) nuclear density was evaluated theoretically and experimentally, and it was revealed that 1 um of thickness change results in about 10% of change in nuclear density.A method to non-destructively measure the thickness of a histopathologic specimen was realized. The method uses the images captured using an optical microscope so that many pathologists can utilize it.Histopathologic images observed using an optical microscope usually have undesirable blur. A method to reduce blur in the 3-D image was realized. The method uses improved blind deconvolution, and satisfactory results were obtained to construct 3-D models of nuclei.A method to estimate nuclear parameters (3-D nuclear density, average and variance of nuclear radii) was realized. The method utilized only a 2-D microscopic image which pathologists are usually observing so that they can easily use the method. Sequential estimation method in which each parameter is estimated sequentially was also developed to reduce the computation time.Finally, a GUI was developed that enables a user to use the developed method easily. The GUI has functions to semi-automatically extract the nuclear positions and nuclear contours, and to estimate nuclear parameters. Nuclear positions and nuclear contours can easily be corrected using simple mouse operations, so pathologists can easily estimate 3-D nuclear density which is independent of the specimen's thickness.