计算机辅助检测在结肠低剂量多层螺旋CT成像中的应用

Feasibility of computer-aided detection for low-dose CT colonography

目的 评估计算机辅助检测系统（CAD）设置不同的检测过滤器数值（SFV）时,在低剂量MSCT成像中对结肠病变的检测能力.方法 56例结肠癌和（或）结肠息肉患者行MSCT结肠成像扫描,依据结肠镜和外科手术结果,将病变分为4组：结肠癌、最长径≥10.0mm息肉、最长径5.1～9.9 mm息肉和最长径≤5.0 mm息肉,之后确定病变在CT图像上的部位及大小,作为评估结肠CAD系统检测病变的金标准.将CAD系统的SFV设为0.25、0.50、0.75和1.00共4个等级,分别检测CT结肠成像图像,记录CAD标注出的病灶的部位和大小,根据上述金标准评估CAD系统对结肠病变的检出率,采用x2检验比较不同SFV设置时CAD对各组病变的检出率.结果 56例患者共有159个阳性病灶,其中结肠癌为44个,最长径≥10.0mm息肉45个,最长径5.1～9.9 mm息肉32个,最长径≤5.0 mm息肉38个.将结肠CAD系统SFV分别设置为0.25、0.50、0.75和1.00时,病灶的检出率分别85.5%（136/159）、85.5%（136/159）、79.2%（126/159）和56.0%（89/159）.SFV为0.25和0.50时,与SFV为1.00时,CAD对病灶检出率的差异有统计学意义（P＜0.05）.随着SFV数值的减低,病灶的检出率增高,假阳性数增加,但91.4%（138/151）～93.9%（31/33）的假阳性病灶很容易识别,仅有6.1%（2/33）～8.6%（13/151）的假阳性病灶,需借助MPR和3D仿真内镜识别.结论 在低剂量MSCT结肠成像中,结肠CAD系统可获得满意的病灶检出率,可调节CAD系统SFV数值,以便满足不同经验阅片者的需求.

Objective To assess the capability of the colon computer-aided-detection（CAD） for the detection of colonic neoplasm on low-dose CT colonography with changing sphericity filter values. Methods Low dose （50 mA） CT colonography data from 56 patients with 159 colonic neoplasms confirmed by conventional colonoscopy and surgery were divided into four groups： colorectal cancer, ≥ 10 mm polyps,5. 1 to 9. 9 mm polyps and ≤ 5.0 mm polyps. The golden standard was established by matching the neoplasms on CT colonography images. The datasets analyzed subsequently by the Colon CAD system at four discrete sphericity filter settings （SFV）： 0. 25, 0. 50, 0. 75 and 1.00. The location and size of highlight lesions by the colon CAD were recorded. Comparison was made with the known golden standard in order to determine the detective rate of colonic neoplasms. Chi-square tests were used to compare the detection rates of four groups by CAD SFV at four settings. Results Fifty six patients had 159 colonic neoplasms including 44 colonic cancers, 45 large polyps （max diameter≥ 10. 0 mm）, 32 median polyps （max diameter 5. 1 to 9. 9 mm） and 38 small polyps （ max diameter ≤5.0 mm）. The colon CAD had the detective rate of 85.5%（ 136/159）, 85.5% （ 136/159 ）, 79. 2% （ 126/159 ） and 56. 0% （89/159） for all colonic neoplasms at four sphericity filter settings 0. 25, 0. 50, 0. 75 and 1.00 respectively. There was statistically significant difference between the sphericity filter settings at 0. 25 and 1.00 （ P ＜ 0. 05）. Detective rate was highest at lower sphericity, false-positive rate increased at same time, however, 91.4% （ 138/151 ）to 93.9% （31/33 ）of false-positives were rapidly dismissible with 2D axial images, only 6. 1% （2/33）to 8.6% （13/151）of false-positives needed to discriminate with multiplanar reconstruction or 3D virtual colonoscopy. Conclusion The Colon CAD system can detect colonic neoplasms satisfactorily in low dose CT colonography and may satisfy the needs of radiologists with different experience by changing sphericity filter settings.