多频域后处理图像对计算机辅助检测数字胸片上肺结节能力的影响

目的评价多频域后处理数字胸片图像对计算机辅助检测系统输出结果的影响。方法选择经CT证实正常248名受检者的DR后前位胸片和30例有肺结节的DR后前位胸片,并应用Philips UNIQUE软件对其进行高频、低频处理。由2名10年以上经验的放射科医生结合CT图像阅读胸片,二人意见达成一致后,在数字化胸片上标记结节位置,保存标记结果并将其作为金标准。然后由CAD系统阅读后处理图像,由2名医师评价CAD系统在后处理图像上肺结节检测的假阳性区域和敏感度。结果 CAD系统在248例低频、常规和高频图像上分别显示了715、585和286个可疑区域,平均每例检出的假阳性区域分别为2.88个、2.36个和1.15个,3组图像检测的假阳性区域及组间差异均具有统计学意义。30例DR图像上共有38个金标准结节,CAD系统在低频、常规和高频图像上分别检测到33、30、28个结节,其检测敏感度分别为86.84%、78.95%和73.68%,ROC曲线下面积分别为0.722、0.670和0.618,3组图像之间检测的差异无统计学意义。结论高频、低频和常规图像模式对CAD系统检测能力无明显影响,但其在低频、常规和高频图像中检测的假阳性有逐渐增高的趋势,肺结节检测敏感度有逐步下降的趋势。

Impact of computer-aided detection system on the nodules of pulmonary indigital chest radiography of image multiscale processing

Objective To evaluate the impact of a available computer-aided detection (CAD) system with an interactive lung nodules analysis software on detecting the pulmonary nodules on digital chest radiograph with image processing. Methods Totally 248 normal subjects and 30 patients with pulmonary nodules confirmed by chest CT were enrolled. Philips UNIQUE software was used to process these images in high-frequency and low frequency. Two experienced chest radiologists (10+ years of experience) read these imaging pictures combined with the relating CT images to mark the location of all nodules with consensus, then stored the diagnostic reports of the marked nodules as golden standard in the CAD. Then CAD system read these images. Two radiologists evaluated the false-positive and sensitivity detection of lung nodules of CAD system on the multiscale processing. Results There were 715, 585 and 286 false-positive nodules respectively on 248 subjects in low frequency, routine frequency and high frequency imagings with an average of 2.88, 2.36 and 1.15 false-positives area (P<0.05). There were 38 lung nodules in 30 digital chest radiographies, the CAD system detected 33, 30 and 28 nodules respectively in low-pass images, standard images and high-pass images, the detection sensitivity was 86.84%, 78.95% and 73.68%, respectively, and the area under the ROC curve was 0.722, 0.670 and 0.618, respectively. Conclusion The prophase experimental results indicate the image multiscale processing of digital chest radiograph does not impact the detectivity of this CAD system, but there are a tendency to increase gradually in detection with false-positive in low-frequency, routine-frequency and high-frequency, and there is a tendency to descend gradually in the detection sensitivity with three multiscale processing images.