Basic study of computer diagnosis of high precision chest x-ray images Using digital radiography

(1)本项目的第一个目标是改进我们开发的AISCR-V3(自动胸片判读系统-版本3)的性能，使其有足够的能力处理最近的数字X线摄影(DR)所拍摄的高质量图像。我们将系统可以处理的图片尺寸从300x300像素扩展到600x600像素。我们开发了一种新的方法来识别血管阴影，使用位置变化的对比度过滤器，提高了对肺部异常阴影的检测能力。(2)关于尘肺的计算机诊断。我们新开发了一种程序来识别肺中的小圆形阴影并测量这些阴影的密度，方法是使用差值类型过滤和Patten分类算法。通过组合不同口罩大小的同一类型过滤器的集合，我们推导出了一种统一进行尘肺密度分类和大小分类的程序。将该方法应用于国际劳工组织国际标准胶片制作中，取得了良好的效果。此外，我们还尝试了尘肺X线CT图像中圆形阴影的识别，并证明了其可行性。(3)两种二值化操作，两幅图像的计算机诊断应用于胸部X线图像。一种是从能量减影DR图像中检测肺癌。这揭示了能量减影图像的特点和计算机解释这类图像的可行性。另一种是自动检测和测量同一患者在不同时间拍摄的图片和序列之间的变化量。提出了一种基于仿射变换的配准和基于局部互相关的变化检测相结合的方法。并将其应用于肺癌的诊断。

(1) Our first aim in this project is to improve the performance of AISCR-V3(Automated Interpretation System of Chest Radiograms - Version 3) which we developed, so that it may have enough ability to treat high quality images taken by recent digital radiography (DR). We extended the picture size the system can process from 300 x 300 pixels to 600 x 600 pixels. We developed a new procedure to recognize vessel shadows using a position variant contast filter, and improved the ability to detect abnormal shfdows in the lung.(2) Concerning computer diagnosis of pneumoconiosis. we newly developed a procedure to recognize small rounded opacities in the lung and measure the density of those opacities, by employing a difference type filtering followed by a patten classification algorithm. By combining the set of the same type of filters with different mask sizes we derived a procedure to execute both the density classification and the size classification of pneumoconiosis in a unified way. We applied the procedure to the ILO international standard film set with the promising result. Furthermore we tried the recognition of rounded opacities from x-ray CT images of pneumoconiosis and showed its feasibility.(3) Two kinds of binary operations two pictures were applied computer diagnosis of chest x-ray images. One is detection of lung cancer from energy subtraction DR images. This revealed characteristics of energy subtraction images and the feasibility of computer interpretation of such images. The other is automated detection and measurement of the amount of changes among picture, sequences taken at different times from the same patient. We developed a procedure consisting of registration by the affine transformation and change detection by the local cross correlation. and applied it to the lung cancer diagnosis.