Development of Automatic Diagnostic System of Pneumoconiosis Using Computed Radiography Chest X-ray Images

随着政府越来越多地参与职业健康维护，需要采取新的医疗决策方法。本课题旨在研制一种基于计算机X线摄影（CR）的尘肺病自动诊断系统。CR系统是一种新型的X射线成像系统，它能产生高质量的X射线图像，其均匀性使开发稳定的自动诊断系统变得容易。该项目的总结如下：（1）已收集了70多个尘肺CR Tilmis。专家读者评价这些档案中尘肺阴影的丰富性。据此建立了尘肺CR档案数据库。(2)已经开发了五种评价混浊度的基本方法。正常/异常分类正确率在70%~ 8%之间。(3)在上述方法的基础上，开发了综合诊断系统。它对正常/异常决策的正确分类率为90%。这个分数是从胸部X线图像的部分区域中获得的，然后我们可以期望总体决策会高得多。这些结果表明，该综合系统是有希望实现自动诊断系统。我们将继续改进每一种方法，并试图达到更高的正确决策率。

The advent of increased government involvement in occupational health maintenance is requiring new approaches to medical decision making. The automatic mass diagnostic screening of medical films for the detection of a specific abnormality could one of promising approaches.This project is to develop an automatic diagnostic system of pneumoconiosis which is based on the Computed Radiography (CR). CR system is a new X-ray imaging system which produce high quality x-ray images and its equality makes it easy to develop stable automatic diagnostic system.The summary of the project is as follows: (1) More than 70 CR Tilmis of pneumoconiosis have been collected. Expert readers evaluated the profusion of opacities of pneumoconiosis of these files. From these results, pneumoconiosis CR file data base has been constructed. (2) Five fundamental methods for evaluating profusion of opacities have been developed. The correct classification rates of them ranges from 70 to 8 percent for normal/abnormal classification. (3) Synthetic diagnostic system has been developed, which is based on the methods mentioned above. Its correct classification rate for normal/abnormal decision is 90 percent. This score is obtained from partial region of chest x-ray images and then we can expect that the overall decision will be much higher.These results show than the synthetic system is promising for the realization of automatic diagnostic system. We will continue to improve each method and try to attain much higher correct decision rate.