DEVELOPMENT OF COMPUTER-AIDED DIAGNOSTIC SYSTEM USING X-RAY IMAGE FEACHER ANALYSIS

本研究从医学x射线图像形成的各个物理机制出发，研究了光学模糊和量子斑驳x射线图像的配准方法。更具体地说，首先利用激光扫描仪的传递特性、薄膜的特性曲线、光密度的对数变换和基于点扩散函数的伪反滤波器确定地消除了光学模糊。其次，根据Bayrs定理推导了两种剩余量子斑的恢复方法。一种是利用量子斑驳的统计特性的随机型宽敏感数字滤波器。另一种是可能缺乏一些信息的新的扩展回归模型和一个简单的实验回归模型。因此，量子斑点在统计上被去除。利用这些滤光片开发并应用于口腔放射学的计算机辅助诊断（CAD）系统。牙本质囊肿、牙源性角化囊肿和成釉细胞瘤多发于颌面部，但其恢复情况各不相同。尽管他们的症状彼此相似，但有必要做出准确的鉴别诊断。作为一项初步研究，我们研究了使用逻辑回归和神经网络的统计方法。在此基础上，我们于1995年7月在日本医学物理学会第12届年会上发表了《基于神经网络和统计分析的口腔放射学鉴别诊断》。并于1995年9月在第36届日本口腔颌面放射学会年会上以“统计学方法与神经网络在牙源性囊肿、牙源性角化囊肿与成釉细胞瘤的影像学鉴别诊断”发表。在这些结果的基础上，我们于1996年构建了系统并积累了数据。1996年9月，Kodera在第37届日本口腔颌面放射学会年会上以“利用专家系统在口腔区域的放射鉴别诊断”为题发表。最后，我们对广岛大学开发的系统进行了临床评价，即牙医对这些方法的有效性进行了评价和确认。将本研究提出的方法应用于牙齿图像的其他区域。少

In this study, some restration methods for the x -ray images with optical blurs and quantum mottles were developed by considering each physical mechanism of the medical x-ray image formation. More specifically, the optical blurs were first deterministically cleared away by using the transfer characteristic of the laser scanner, the characteristic curve of the film, the logarithmic transformation of the optical density and a pseud-invers filter based on the point spread function. Next, two restoration methods for remaining quantum mottles were derived from the basis of a Bayrs' theorem. One was a wide sensive digital filter of a stochastic type using the statistical properties of quantum mottles. Another was a new extended regression model that might have a little information lack and a simple experimental regression model. Thus, the quantum mottles were statistically removed. A computer-aided diagnostic (CAD) system using these filters is developed and applied for oral radiology. Denti … More gerous cyst, odontogenic keratocyst and ameloblastoma in maxillofacial region occur frequently, but their recuperations are diffrent. Then it is necessary to make an exact differential diagnosis, although their symptoms are similar to one another. As a pilot study, we investigated statistical methods using a logistic regression and neural network. Under these studies, we presented as "Differential diagnosis by neural network and statistical analysis in oral radiology" in the 12th Annual eeting Japanese Association of Medical Physics in July 1995. And, we also presented as "Radiographic differential diagnosis among dentigerous cyst, odontogenic keratocyst and ameloblastma using statistical methods and neural networrk" in the 36th Japanese Society for Oral and Maxillofacial Radiology Annual meeting in September 1995. Based on these results, in 1996, we constructed the system and accumulated data. Kodera presented as "Radiographic differentiial diagnosis in dental region using expert system" in teh 37th Japanese Society for Oral and Maxillofacial Radiology Annual meeting in September 1996. Finally, we evaluated clinically the developed system in HIroshima University, namely dentists evaluated and confirmed the deffectiveness of these methods. The method proposed in this study was applied to the other area in dental images. Less