Megibow Alec J, Rusinek Henry, Lisi Virna, Bennett Genevieve L, Macari Michael, Israel Gary M, Krinsky Glenn A
Department of Radiology, NYU Medical Center, New York, NY 10016, USA.
J Digit Imaging. 2002 Jun;15(2):84-90. doi: 10.1007/s10278-002-0009-4. Epub 2002 Sep 26.
Using receiver-operating characteristic (ROC) methodology, the ability to diagnose acute appendicitis with computed tomography (CT) images displayed at varying levels of lossy compression was evaluated. Nine sequential images over the ileocecal region were obtained from 53 consecutive patients with right lower quadrant pain who were clinically suspected to have acute appendicitis. Thirty were proven surgically to have acute appendicitis, alternative diagnoses confirmed in 23. The image sets were subjected to a lossy wavelet-based compression algorithm "Embedded Predictive Wavelet Image Coder" (EPWIC). Compression levels were: none, 8:1, 16:1, and 24:1, resulting in 4 sets of images per patient. Image sets were randomized and evaluated separately by 4 body radiologists on a 1,024 x 768-pixel SVGA color PC monitor in 512 x 512 format. The readers were aware of the clinical suspicion of appendicitis but were unaware of the positive fraction of cases. Individual and combined reader ROC and c2 analyses of sensitivity, specificity, and accuracy were determined. For all readers, sensitivity decreases at 16:1 and 24:1 levels (P <0.01, P <0.001, respectively). Accuracy decreased at 24:1 levels (P <0.01). Specificity was unaffected. By ROC analysis there was statistically significantly decreased area under the curve at 24:1 levels (P <0.02) as compared with uncompressed images. Finite levels of lossy wavelet compression may be applied to CT images without compromising diagnostic performance.
采用接受者操作特征（ROC）方法，评估了在不同有损压缩水平下利用计算机断层扫描（CT）图像诊断急性阑尾炎的能力。从53例临床怀疑患有急性阑尾炎且右下象限疼痛的连续患者中获取了9张回盲部区域的连续图像。其中30例经手术证实患有急性阑尾炎，23例确诊为其他疾病。图像集采用基于小波的有损压缩算法“嵌入式预测小波图像编码器”（EPWIC）进行处理。压缩水平分别为：无压缩、8:1、16:1和24:1，每位患者由此得到4组图像。图像集随机分组后，由4位身体放射科医生在一台1024×768像素的SVGA彩色PC显示器上以512×512格式分别进行评估。阅片者知晓临床对阑尾炎的怀疑，但不知晓病例的阳性比例。确定了阅片者个体及联合的ROC以及敏感性、特异性和准确性的卡方分析结果。对于所有阅片者，在压缩比为16:1和24:1时敏感性均降低（P分别<0.01和<0.001）。在压缩比为24:1时准确性降低（P<0.01）。特异性不受影响。通过ROC分析，与未压缩图像相比，在压缩比为24:1时曲线下面积有统计学显著下降（P<0.02）。有限水平的小波有损压缩可应用于CT图像而不影响诊断性能。
