Recognition of the pulmonary vessels by knowledge base of the radiologic anatomy of the bronchopulmonary tree for the computer diagnosis of the tumor of the lung.

在胸片或胸部CT周围肺结节影自动诊断系统中，假阳性影的鉴别至关重要。在计算机分析系统的图像识别研究中，肿瘤的结节阴影显示的输出值与胸片上血管阴影的输出值相似。使用专门设计的提取滤波器和在适当阈值水平上消除的分析技术难以识别这些图像。构建基于三维解剖的医学图像知识库是本研究的目的，旨在建立肺结节的检测方法和假阳性阴影的消除方法。【方法与结果】1)基于三维解剖学的图像知识库最初建立了包含血管长度和血管方向信息的解剖学知识库。它包括从门部到外周血管的信息，更能反映血管的原始分支模式。本研究采用Yamashita创建的流行模式组合作为知识库模型。当右主支气管的长度被设定为三个单位时，血管的长度被确定。方向被三维显示为26个方向之一。血管的连续性表现为亲子-孙辈关系。2)基于解剖知识的血管决策基于具有长度、方向和连续性信息的解剖知识库构建血管决策算法。然而，仅凭这个解剖学知识数据库是不可能命名血管的。3)肺结节自动检测系统利用胸部螺旋CT图像，研制了肺结节自动检测系统。为了检测CT图像上的结节，我们使用了针对胸片开发的结节定向对比滤波器（DCF-N），并对其参数进行了修改。该系统成功检出了87.5%的结节，但也检出了许多假阳性灶。包括血管解剖信息，将有可能减少假阳性，提高该系统的鉴别诊断价值。[结论]该系统识别血管的能力可能不足以用于未来的应用。造成这种情况的原因之一可能是由于与动脉树的多个分支模式模型相对应的信息不足。肺动脉与静脉的长度、方向、连续性等信息难以区分。在系统中必须定义船舶的位置信息。在结节检测方面，该系统能较好地检测出具有三同心圆形结构的结节性混浊物。少

Discrimination of false positive shadow is important in an automatic diagnosis system for nodular shadow of the peripheral lung on chest radiogram or on chest CT.In an image recognition study by computed analysis system, nodular shadow of the cancer revealed an output value that looks like those of the shadows of the blood vessels on the chest radiogram. It is difficult to identify those images by analytical technique using specially designed extraction filter and elimination at an appropriate threshold level. Construction of the knowledge database of medical images based on the 3-dimensional anatomy is an aim of this study to establish a detection method for pulmonary nodules and an elimination method for false positive shadows. [Methods and Results]1)Kowledge database of images based on the 3-dimensional anatomyAn anatomical knowledge base which has length and information of direction of blood vessels was originally created. It includes information from the hilum to the periphery ref … More lecting original branching pattern of the blood vessels. As a model of knowledge base, a combination of prevalence patterns that created by Yamashita was used in this study. Length of the vessels were decided as the length of the right main bronchus was set to three units. Directions was 3-dimensionally demonstrated as one of 26 directions. Continuity of the vessels was shown as parent-children-grandchildren relationships.2)Decision of blood vessels on the basis of anatomical knowledge baseDecision algorithm of the blood vessels on the basis of anatomical knowledge base which has information of length, direction, and continuity was created. However, it was impossible to name vessels on only this anatomical knowledge database.3)Automated detection system for pulmonary nodulesUsing chest helical CT image, an automated detection system for pulmonary nodules has been developed. To detect nodules on CT images, we use the directional contrast filter for nodule (DCF-N) which was developed for chest radiograms and its parameters were modified for CT images. Nodules of 87.5% was successfully detected by this system, however, many falsepositive foci was detected by this system. Including information of vascular anatomy, it will be possible to reduce false-positives and improve the differential diagnostic value of this system. [Conclusions]The ability of this system to recognition blood vessels may not be sufficient for future use. One of the reason of this situation might be due to insufficient information that correspond to multiple branching pattern models of the arterial tree. There was a difficulty to differentiate pulmonary arteries from veins by the information of length, direction, and continuity. Information of lobar position of the vessels must be defined in the system. As for nodule detection, the system fairly detected nodular opacities which have three-concentric circular structures. Less