MRI断层层厚测量方法研究

Study of fast and accurate methods for testing slice thicknesses of magnetic resonance images

探讨客观、快速、准确测量二维磁共振图像断层层厚的方法。扫描有层厚测试功能的体模获得相应图像,利用传统测试方法(方法 1:调节窗宽、窗位测量目标物的方法;方法 2:单条层面灵敏度剖面线的方法)与改进的多条灵敏度剖面线平均(方法 3A)及一阶导处理自动搜寻目标物边缘的方法(方法 3B)测量层厚。标称层厚为5 mm,当图像均匀性较好时(98.97%),得到各个方法的层厚均值和标准偏差,方法1、2、3A、3B结果分别为4.96 mm/0.03 mm,4.97 mm/0.06 mm,4.98 mm/0.02 mm,4.97 mm/0.01 mm,单因素方差分析F值为0.48;当图像均匀性较差时(95.35%),方法1、2、3A、3B结果分别为4.82 mm/0.41 mm,4.80 mm/0.46 mm,4.84 mm/0.15 mm,4.85 mm/0.11 mm,F值为0.46。F值结果显示4种层厚测试方法有效,但是标准偏差的差异意味着测试结果的不确定度变化较大。方法 1标准偏差最大,方法3B的标准偏差最小。4种测试方法的结果表明:传统测试方法主观性强,误差大,不确定度大,稳定性欠佳;改进后的方法客观、快速,准确度高。特别是当图像均匀性欠佳时,改进的层厚测试方法抗干扰性强。

The paper explores an objective, fast and accurate method for testing the slice thicknesses of 2D magnetic resonance images( MRI). The phantom with slice thickness testing module was scanned to acquire corresponding MRI. The traditional slice thickness testing methods( Method 1:Adjusting the window level and window width to measure the target object; Method 2:Single slice profile method), the improved method which applied the average of multiple profiles( Method 3A)and the first-order derivative for searching the boundary of the target object( Method 3B) had been used.As the image uniformity was as high as( 98.97%), the slice thicknesses and the standard deviations obtained through the methods 1, 2, 3A and 3B were 4.96 mm/0.03 mm, 4.97 mm/0.06 mm,4.98 mm/0.02 mm and 4.97 mm/0.01 mm, respectively; the one-way ANOVA value F was 0.48. As the image uniformity was lowered to( 95.35%), the slice thicknesses and the standard deviations obtained through the methods 1, 2, 3A and 3B were 4.82 mm/0.41 mm, 4.80 mm/0.46 mm,4.84 mm/0.15 mm and 4.85 mm/0.11 mm respectively; F was 0.46. The F value indicates that the above testing methods are effective, but the difference in standard deviations reveals that the uncertainty( U) in testing results stated above changes greatly. The standard deviation got from Method 1 is the largest and that of Method 3B is the smallest. The testing results introduced in the paper show that traditional methods are too subjective and unstable with big errors and large uncertainty, while the improved methods stated above are objective, fast and accurate, and more particularly, strong enough to resist interference in case of poor image uniformity.