Intelligent CAD Based on Anatomical Classification

最近，在日本，医生诊断大量多探测器行CT图像是一个很大的负担。我们的项目旨在开发一个复杂的系统，为医生诊断大量的CT图像有效地通过CAD。本研究开发的主要项目如下：-建立多器官、多疾病的海量图像数据库。-基于三维CT图像的多疾病检测的发展。开发利用三维CT图像辅助诊断肺部疾病、心血管疾病、骨骼疾病的综合系统。本研究的主要目标疾病为肺癌、肺气肿、冠状动脉钙化、骨质疏松症。研究结果如下：1.多器官、多疾病的海量影像数据库的建设：在医学信息的二次利用中，必须得到各医疗机构伦理委员会的批准。委员会要求保护 ...更多信息 DICOM标题信息的个人信息的删除。我们开发了一种匿名化方法，可以灵活地遵守每个医疗站点指定的策略，然后，开发了一个系统，在每个医疗站点的大规模DICOM图像平滑匿名。该系统目前在五个医疗点运行。基于三维CT图像的多疾病检测的研究进展(2-1)多器官分割和定量分析：我们开发了可以分割解剖结构的方法，如骨骼、胸壁、纵隔、横膈、血管、支气管等。引入与胸部解剖相关的空间位置，开发了支气管命名和肺叶分割技术。(2-2)肺癌候选者的检测：我们使用低剂量多探测器行CT图像，基于从大病例到小病例的结节候选者的肺解剖信息开发了一种检测算法。(2-3)肺气肿候选者的检测：我们开发了一种方法来提取低剂量CT三维图像上肺部区域的低衰减区（LAA）。检测方法允许我们分析检测到的LAA的体积和分布模式。此外，可以跟踪LAA容积的时间间隔变化。(2-4)冠状动脉钙化候选检测：提出了一种自动识别冠状动脉高衰减区域的方法。为了提高检测的准确性，我们使用分割的主动脉和肺动脉区域来减少检测区域。(2-5)骨质疏松症候选者的检测：我们开发了一种胸椎提取算法来测量松质组织内的CT值和骨密度。通过比较在生成的基础上提取的松质组织内部的平均CT值，我们开发了一种区分异常和正常骨的算法.集成系统：我们一直在集成的检测算法到一个原型系统与图形用户界面。我们验证了原型系统的有效性，使用大型数据集与医学专家的合作。少

Recently, it is a large amount of burden for physicians to diagnose massive multi-detector row CT images in Japan. Our project aims to develop a sophisticated system for physicians to diagnose a large amount of the CT images effectively through CAD. Principal items of this research and development are as follows:-Construction of massive image database of multi-organ, multi-disease.-Development of multi-diseases detection based on 3-D CT images.-Development of integrated system that aids physicians to diagnose lung diseases, heart vessel diseases, and bone diseases using 3-D CT images.In this study, main target diseases for detection are lung cancer, pulmonary emphysema, calcification of coronary artery, and osteoporosis. Research results are as follows.1. Construction of massive image database of multi-organ, multi-disease: In secondary use of medical information for research and education, approval by ethics committee in each medical site is essential. The committee requires the prote … More ction of personal information of DICOM header information. We developed anonymization method which can flexibly comply with the policy that specified by each medical site and then, developed a system which smoothly anonymizes large-scale DICOM images in each medical site. The system is now operating in five medical site.2. Development of multi-diseases detection based on 3-D CT images.(2-1) multi-organ segmentation and quantitative analyses: We developed methods that can segment anatomical structures such as bone, chest wall, mediastinum, diaphragm, vessel, bronchus, etc. Introducing spatial location concerning the thoracic anatomy, bronchus nomenclation and lung lobe segmentation techniques were developed.(2-2) Detection of candidate of lung cancer: We developed a detection algorithm based on the lung anatomical information of nodule candidates from large cases to small cases with GGO using low-dose multi-detector row CT images.(2-3) Detection of candidate of pulmonary emphysema: We developed a method to extract low attenuation areas (LAA) in lung region on 3-D low-dose CT images. The detection method allows us to analyze volume and distribution patterns of the detected LAA. Additionally, it is possible to trace the time interval change of LAA volume.(2-4) Detection of calcification candidate of coronary artery: We developed a method to extract high attenuation areas on coronary artery that is automatically identified. In order to improve the detection accuracy, we reduced detection area by using the segmented aorta and pulmonary artery region.(2-5) Detection of candidate of osteoporosis: We developed an extraction algorithm of thoracic vertebrae to measure CT number inside cancellous tissue and bone density. From the comparison of the average CT value inside the extracted cancellous tissue on the basis of the generation, we developed an algorithm for differentiating abnormalities from normal bones.3. Integrated system: We have been integrating the detection algorithms into a prototype system with graphical user interface. We validated the effectiveness of the prototype system using a large dataset with the cooperation of medical experts. Less