基于双能减影信息的骨骼抑制算法研究

When lung areas overlap with bones in chest radiography, it would affect the earle-stage detection of malignant lund disease, especially for lung cancer. Even though a dual-energy subtraction provides more detail of lesion without shelter of bones, patients will be exposed to additional radiation dose. An algorithm is proposed for modeling bones based on the given dual-energy subtraction information and suppressing the contrast of bones in chest radiographs to overcome these two problems. The produced soft-tissue images without the effect of bones would be employed for the early screening of lung disease, and would improve the sensitivity and accuracy of detection.. In this study, the lung segmentation method, based on the gray and shape information in the features images, is employed to remove the areas outside of the lung. And third-order B-spline wavelet transform for multi-scale analysis and the extraction of local 2-jet feature are used to represent bones. Then, support vector regression based on the dual-energy subtraction is employed to build the models of bones, which is employed to predict the gray level distribution of bones in chest radiography. The result of prediction is corrected by analysis of survival probability, and processed by a difference-of-Gaussians filter. Finally, a subtraction technique is employed to suppress the contrast of bones, and produce a soft-tissue image without bones. Our study would not only achieve the result in the field of medical image processing and analysis, but also be employed for the early screening of lung disease, which would improve the performance of diagnosis of malignant lung disease. .

胸部X光图像中的肺组织与骨骼区域重叠影响肺癌等恶性肺部疾病的早期检测。而双能减影图像虽能提供无骨骼遮挡的病灶显示，但却显著增加患者所受放射线剂量。针对这两种成像问题，本项目提出利用已有双能减影信息建立骨骼模型，实现胸部X光图像骨骼抑制的算法。将生成无骨骼遮挡的软组织图像应用于肺部疾病的早期筛查，提高检测的敏感度与准确性。.本项目先利用基于特征图像灰度与形状信息的肺区域分割算法去除肺外无关区域，再利用多尺度三阶B样条小波变换及局部2-jet特征提取来有效地表示骨骼结构。然后，使用基于双能减影信息的支持向量回归算法构建骨骼模型，预测胸部X光图像的骨骼灰度分布，并利用可信概率分析及高斯差值滤波等对预测结果进行修正。最后，利用减影技术实现骨骼抑制，生成无骨骼遮挡的软组织图像。本项目研究不仅将在医学图像处理与分析领域取得理论研究成果，还将应用于肺部疾病的早期筛查，提高肺癌等恶性疾病的临床诊断性能。

胸部X光图像中肺组织与骨骼区域重叠影响肺癌等恶性肺部疾病的早期检测。而双能减影图像虽能提供无骨骼遮挡的病灶显示，但却显著增加患者所受放射线剂量。针对这两种成像问题，本项目利用已有双能减影信息建立骨骼模型实现胸部X光图像骨骼抑制。将生成无骨骼遮挡的软组织图像应用于肺部疾病的早期筛查，提高检测的敏感度与准确性。.本项目先利用基于先验模型的肺区域分割算法去除肺外无关区域，再利用多尺度三阶B样条小波变换及局部2-jet特征提取来有效地表示骨骼结构。然后，使用基于双能减影信息的支持向量回归算法构建骨骼模型，预测胸部X光图像中骨骼的灰度分布，并利用可信概率分析及高斯差值滤波等对预测结果进行修正。最后，利用减影技术实现骨骼抑制，生成无骨骼遮挡的软组织图像。.在本项目中，使用双能减影图像中的软组织图像与得到的抑制肋骨结果进行比较。实验数据为沈阳某大型医院放射科提供的200组胸部双能减影图像。每组包括患者的三张图像，分别为正常胸部X光图像、骨骼图像及软组织图像。成像设备为GE Definium6000双能DR成像系统。.本项目所研究方法取得令人满意的较好性能。其研究不仅将在医学图像处理与分析领域取得理论研究成果，还将应用于肺部疾病的早期筛查，提高肺癌等恶性疾病的临床诊断性能。