Erdi Y E, Wessels B W, Loew M H, Erdi A K
Department of Radiation Oncology and Biophysics, George Washington University, Washington, DC 20037, USA.
Cancer Res. 1995 Dec 1;55(23 Suppl):5823s-5826s.
Thresholding is the most widely used organ or tumor segmentation technique used in single photon emission computed tomography (SPECT) and planar imaging for monoclonal antibodies. Selecting the optimal threshold requires a priori knowledge (volumes from CT or magnetic resonance) for the size and contrast level of the organ in question. Failure to select an optimal threshold leads to overestimation or underestimation of the volume and, subsequently, the organ-absorbed dose value in radio-immunotherapy. To investigate this threshold selection problem, we performed a phantom experiment using six lucite spheres ranging from 1 to 117 ml and filled with a uniform activity of 1 microCi/ml Tc-99m. These spheres were placed at the center and off-center locations of a Jasczsak phantom and scanned with a three-headed gamma camera in SPECT and planar modes. Target-nontarget (T:NT) ratios were changed by adding the appropriate activity to the background. A threshold search algorithm with an interpolative background correction was applied to sphere images. This algorithm selects a threshold that minimizes the difference between the true and measured volumes (SPECT) or areas (planar). It was found that for spheres equal to or larger than 20 ml [diameter (D) > 38 mm] and T:NT ratios higher than 5:1, mean thresholds at 42% for SPECT and 38% for planar imaging yielded minimum image segmentation errors, which is in agreement with current literature. However, for small T:NT ratios (< 5:1), the threshold values as high as 71% for SPECT and 85% for planar imaging were substantially different than those fixed thresholds for large spheres (D > 38 mm). Hence, the use of fixed thresholds in low contrasts and with tumor and organ sizes of clinical interest (25 < or = D < or = 50 mm) may result in limited volume estimation accuracy. Therefore, we have provided the investigator a method to obtain the threshold values in which the proper threshold can be selected based on the organ and tumor size and image contrast. By measuring and calibrating the proper threshold value derived through machine-specific phantom measurements, a more accurate volume and activity quantitation can be performed. This, in turn, will provide tumor-absorbed dose optimization and greater accuracy in the measurement of potentially subacute, toxic absorbed doses to normal organs for patients undergoing radioimmunotherapy.
阈值分割是单光子发射计算机断层扫描（SPECT）和单克隆抗体平面成像中使用最广泛的器官或肿瘤分割技术。选择最佳阈值需要先验知识（来自CT或磁共振的体积数据），以了解所研究器官的大小和对比度水平。未能选择最佳阈值会导致对体积的高估或低估，进而导致放射免疫治疗中器官吸收剂量值的偏差。为了研究这个阈值选择问题，我们进行了一个体模实验，使用了六个体积从1到117毫升的有机玻璃球，每个球填充有均匀活度为1微居里/毫升的Tc-99m。这些球被放置在Jasczsak体模的中心和偏心位置，并使用三头伽马相机以SPECT和平面模式进行扫描。通过向背景添加适当的活度来改变靶-非靶（T:NT）比值。对球体图像应用了一种带有插值背景校正的阈值搜索算法。该算法选择一个能使真实体积与测量体积（SPECT）或面积（平面）之间的差异最小化的阈值。结果发现，对于等于或大于20毫升[直径（D）> 38毫米]且T:NT比值高于5:1的球体，SPECT的平均阈值为42%，平面成像的平均阈值为38%时，图像分割误差最小，这与当前文献一致。然而，对于小的T:NT比值（< 5:1），SPECT的阈值高达71%，平面成像的阈值高达85%，与大球体（D > 38毫米）的固定阈值有很大不同。因此，在低对比度以及临床关注的肿瘤和器官大小（25≤D≤50毫米）情况下使用固定阈值可能会导致体积估计精度有限。所以，我们为研究人员提供了一种获取阈值的方法，可根据器官和肿瘤大小以及图像对比度来选择合适的阈值。通过测量和校准通过特定机器的体模测量得出的合适阈值，可以进行更准确的体积和活度定量。这反过来将为接受放射免疫治疗的患者提供肿瘤吸收剂量优化，并在测量对正常器官潜在的亚急性毒性吸收剂量时提高准确性。
