目的分析基于从头训练模式深度学习-卷积神经网络模型[the deep learning convolutional neural network model trained from scratch,DL-CNN(fs)]的人工智能算法评估急性肺动脉血栓栓塞(acute pulmonary thromboembolism,APE)的价值。...目的分析基于从头训练模式深度学习-卷积神经网络模型[the deep learning convolutional neural network model trained from scratch,DL-CNN(fs)]的人工智能算法评估急性肺动脉血栓栓塞(acute pulmonary thromboembolism,APE)的价值。方法回顾性纳入214例可疑APE行CT肺动脉造影(CTPA)的住院患者,包括急性肺动脉血栓栓塞137例,阴性77例。放射科医师根据CTPA图像判断有无APE,并计算Qanadli评分、Mastora评分和其他CTPA参数。采用DL-CNN(fs)训练网络模型自动检测栓子的分布及容积。评估DL-CNN(fs)模型测量血栓分布的价值,计算血栓负荷与Qanadli评分、Mastora评分和其他CTPA参数的相关性。结果DL-CNN(fs)测算的中心肺动脉栓子敏感度、特异度、感兴趣区曲线下面积(AUC)分别为100%、16.8%、0.584(95%CI,0.508~0.661);DL-CNN(fs)测算的外周肺动脉栓子敏感度、特异度、AUC均较高(R1-R9,60.8%~95.2%,67.9%~87.1%,0.740~0.844;L1-L10,64.6%~93.4%,62.7%~83.1%,0.732~0.791)。DL-CNN(fs)测算的栓子体积与Qanadli score肺栓塞指数显著正相关(r=0.867,P<0.001),与Mastora score肺栓塞指数显著正相关(r=0.854,P<0.001),与右心室及左心室最大横径比、右心室及左心室最大面积比呈正相关(r=0.549,0.559,P<0.01)。结论DL-CNN(fs)模型检测外周肺动脉栓子具有较高的价值,对中心肺动脉栓子诊断特异度有待进一步提高。DL-CNN(fs)模型自动提供APE患者的栓子体积,可以一定程度反映栓塞程度及右心功能,能够辅助医生对于APE患者血栓负荷及危险分层的快速评估。展开更多
AIM: To assess lung parenchymal changes in ankylosing spondylitis (AS) using high resolution computed tomography (HRCT). METHODS: We included 78 AS patients whose average age was 33.87 (18-56) years with a ratio of 53...AIM: To assess lung parenchymal changes in ankylosing spondylitis (AS) using high resolution computed tomography (HRCT). METHODS: We included 78 AS patients whose average age was 33.87 (18-56) years with a ratio of 53 males to 25 females who were followed up for 3.88 (1-22) years on average. neumonia and tuberculosis were excluded. In a detailed examination of lung HRCT findings, we investigated the presence of parenchymal micronodules,parenchymal bands, subpleural bands, interlobular and intralobular septal thickening, irregularity of interfaces,ground glass opacity, consolidation, mosaic pattern,bronchial wall thickening, bronchial dilatation, tracheal dilatation, pleural thickening, emphysema, thoracic cage asymmetry, honeycomb appearance, structural distortion, apical fibrosis and other additional findings.RESULTS: In detailed HRCT evaluations, lung parenchymal changes were found in 46 (59%) of all patients. We found parenchymal bands in 21 (27%) cases, interlobular septal thickening in 9 (12%), emphysema in 9 (12%), apical fibrosis in 8 (10%), ground-glass opacities in 7 (9%), parenchymal micronodules in 5 (6%), irregularity in interfaces in 3 (4%), bronchial dilatation in 3 (4%), mosaic pattern in 2 (3%), pleural thickening in 2 (3%), consolidation in 1 (1%), bronchial wall thick ening in 1 (1%) and a subpleural band in 1 (1%) case. Furthermore, we detected subsegmental atelectasis in 2 patients and a cavitary lesion in 1 patient. CONCLUSION: Our study had the highest number of AS cases of all previous studies in evaluating lung paren chymal changes. The rate of lung parenchymal changes was slightly lower than that reported in recent literature.展开更多
文摘目的分析基于从头训练模式深度学习-卷积神经网络模型[the deep learning convolutional neural network model trained from scratch,DL-CNN(fs)]的人工智能算法评估急性肺动脉血栓栓塞(acute pulmonary thromboembolism,APE)的价值。方法回顾性纳入214例可疑APE行CT肺动脉造影(CTPA)的住院患者,包括急性肺动脉血栓栓塞137例,阴性77例。放射科医师根据CTPA图像判断有无APE,并计算Qanadli评分、Mastora评分和其他CTPA参数。采用DL-CNN(fs)训练网络模型自动检测栓子的分布及容积。评估DL-CNN(fs)模型测量血栓分布的价值,计算血栓负荷与Qanadli评分、Mastora评分和其他CTPA参数的相关性。结果DL-CNN(fs)测算的中心肺动脉栓子敏感度、特异度、感兴趣区曲线下面积(AUC)分别为100%、16.8%、0.584(95%CI,0.508~0.661);DL-CNN(fs)测算的外周肺动脉栓子敏感度、特异度、AUC均较高(R1-R9,60.8%~95.2%,67.9%~87.1%,0.740~0.844;L1-L10,64.6%~93.4%,62.7%~83.1%,0.732~0.791)。DL-CNN(fs)测算的栓子体积与Qanadli score肺栓塞指数显著正相关(r=0.867,P<0.001),与Mastora score肺栓塞指数显著正相关(r=0.854,P<0.001),与右心室及左心室最大横径比、右心室及左心室最大面积比呈正相关(r=0.549,0.559,P<0.01)。结论DL-CNN(fs)模型检测外周肺动脉栓子具有较高的价值,对中心肺动脉栓子诊断特异度有待进一步提高。DL-CNN(fs)模型自动提供APE患者的栓子体积,可以一定程度反映栓塞程度及右心功能,能够辅助医生对于APE患者血栓负荷及危险分层的快速评估。
基金funded by the 12th Five-year National Plan for Science and Technology(2012BAK16B02)the Council of National Science Foundation of China(81571851,81401559)+1 种基金the Scientific and Technological Key Project of Shanghai Municipality(14231202500,14DZ2271500)the Central Research Institute Public Project(GY2013Z-3,GY2014Z-1)
文摘AIM: To assess lung parenchymal changes in ankylosing spondylitis (AS) using high resolution computed tomography (HRCT). METHODS: We included 78 AS patients whose average age was 33.87 (18-56) years with a ratio of 53 males to 25 females who were followed up for 3.88 (1-22) years on average. neumonia and tuberculosis were excluded. In a detailed examination of lung HRCT findings, we investigated the presence of parenchymal micronodules,parenchymal bands, subpleural bands, interlobular and intralobular septal thickening, irregularity of interfaces,ground glass opacity, consolidation, mosaic pattern,bronchial wall thickening, bronchial dilatation, tracheal dilatation, pleural thickening, emphysema, thoracic cage asymmetry, honeycomb appearance, structural distortion, apical fibrosis and other additional findings.RESULTS: In detailed HRCT evaluations, lung parenchymal changes were found in 46 (59%) of all patients. We found parenchymal bands in 21 (27%) cases, interlobular septal thickening in 9 (12%), emphysema in 9 (12%), apical fibrosis in 8 (10%), ground-glass opacities in 7 (9%), parenchymal micronodules in 5 (6%), irregularity in interfaces in 3 (4%), bronchial dilatation in 3 (4%), mosaic pattern in 2 (3%), pleural thickening in 2 (3%), consolidation in 1 (1%), bronchial wall thick ening in 1 (1%) and a subpleural band in 1 (1%) case. Furthermore, we detected subsegmental atelectasis in 2 patients and a cavitary lesion in 1 patient. CONCLUSION: Our study had the highest number of AS cases of all previous studies in evaluating lung paren chymal changes. The rate of lung parenchymal changes was slightly lower than that reported in recent literature.
