基于深度学习的胸部数字X线质量控制方法  

Quality control method of chest digital radiography based on deep learning

作　　者：易音巧 王一达 宋阳[1] 杨永贵 汪劭川 郭岗 杨光[1] YI Yinqiao;WANG Yida;SONG Yang;YANG Yonggui;WANG Shaochuan;GUO Gang;YANG Guang(Shanghai Key Laboratory of Magnetic Resonance(East China Normal University),Shanghai 200062,China;Department of Radiology Imaging,The 2nd Affiliated Hospital of Xiamen Medical College,Xiamen Fujian 361000,China;Xiamen Jiekang Intelligent Medical Technology Company Limited,Xiamen Fujian 361000,China)

机构地区：[1]上海市磁共振重点实验室(华东师范大学),上海200062 [2]厦门医学院附属第二医院放射影像科,福建厦门361000 [3]厦门捷康智慧医疗科技有限公司,福建厦门361000

出　　处：《计算机应用》2021年第S01期237-242,共6页journal of Computer Applications

基　　金：国家自然科学基金重点项目(61731009)。

摘　　要：针对人体摆位姿势影响胸部数字X线成像(DR)质量的这一问题,提出了一种基于深度学习的质量控制方法。首先通过YOLOv3模型实现影响DR胸片图像质量的肩胛骨区域的自动定位;然后,将检测到的肩胛骨区域输入到基于残差网络的深度学习模型对肩胛骨是否推出肋骨外进行自动判断;最后,根据网络输出的结果指导拍片技师进行相应的调整,拍摄质量合格的胸部数字X线成像。实验结果表明,使用YOLOv3模型进行肩胛骨位置检测时,当交并比(IoU)设为0.8时,右侧肩胛骨检测的平均精确率(AP)为98.75%,左侧肩胛骨检测的AP为98.86%,检测结果精确,可用于后续的分类模型。分类网络利用左右肩胛骨区域图像进行分类,在测试集上,接受者操作特性曲线(ROC)下与坐标轴围成的面积(AUC)为0.855。当阈值取0.45时,约旦指数最大,此时准确率为0.789,敏感性为0.801,特异性为0.783,阳性预测率为0.672,阴性预测率为0.876。由此可见,所提出方法可以有效地对肩胛骨是否推出肋骨外进行自动判断,辅助医生对胸部数字X线的图像质量进行控制。In order to address the influence of the posture of patients on the quality of the Digital Radiography(DR)images,a quality control method based on deep learning was proposed.Firstly,YOLOv3 model was used to realize the automatic positioning of the scapula in the chest DR image.Then the deep learning classification network model based on the residual network was used to judge whether the scapula was pushed out of the ribs.Finally,according to the result of whether the scapula was out of the ribs output from the network,the technician was instructed to make corresponding adjustments to shoot qualified chest DR image.The experiment results showed that the trained YOLOv3 model could detect the scapula regions accurately.When the Intersection over Union(IoU)was set to 0.8,the Average Precisions(APs)of the right and left scapulae achieved 98.75%and 98.86%respectively.When the classification model was used to determine whether the scapulae regions were pushed out of ribs,the Area enclosed by the coordinate axis Under the ROC(Receiver Operating Characteristic)curve(AUC)is 0.855 over the test dataset.When the threshold was set at 0.45 according to the maximum Jordan index,the model achieved an accuracy of 0.789,a sensitivity of 0.801,a specificity of 0.783,a positive predictive value of 0.672,and a negative predictive value of 0.876.It can be seen that the proposed method can effectively determine whether the scapulae were pushed out of the ribs and can assist the doctor to control the image quality of the chest DR image.

关 键 词：深度学习 数字X线成像 迁移学习 物体检测 图像分类 质量评估 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]