Image guided radiotherapy (IGRT) provides the basis for precise treatment of intensity-modulated radiotherapy, and kV X-ray imaging is used frequently during a course of radiation therapy to improve the precision and accuracy of the delivery of the treatment of lung cancer [
16]. IGRT can verify the treatment position prior to the implementation of the treatment, measure and analyze the three-dimensional error of the central position of tumors and correct it online, which has become one of the important bases for the implementation of precision therapy [
17]. In the system of IGRT, CBCT has attracted significant attention from practitioners who seek to enhance diagnosis and treatment for their patients because it provides multidimensional and dimensionally accurate images for diagnosis and treatment planning [
18] However, CBCT is a conical beam scanning, and has a certain gap between fan Beam CT in the image clarity, especially for lung tumors with obvious respiration problems [
19]. Therefore, how to quickly and accurately register CBCT images with planned images, especially for chest tumors that are easily affected by respiration, needs a further investigation. Lung cancer is the leading cause of cancer deaths worldwide especially in developing countries [
20]. In this study, we evaluated the accuracy of CBCT image registration used in image-guided radiotherapy of lung cancer.
Image registration with tumor as marker point was first put into CBCT clinical application in a previous study. If there is a significant difference in density between tumor and surrounding lung tissue, the tumor itself has the basic conditions as a registration marker [
21]. However, due to the long time it takes to obtain CBCT in practice, the tumor will move with breathing [
22], and the obtained CBCT image includes information about breathing movement, thus the accuracy of image matching is obviously affected. Moreover, with the treatment, the tumor is likely to shrink, and shows obvious difference with planned image of the tumor, leading to the difficulty of registration [
23]. In the present study, anatomical structures such as vertebral body, protrusion, etc. were used because they are significantly different compared to the density of the surrounding tissues and do not change significantly with the passage of treatment. The analysis of 160 sets of registration data of 8 patients with lung cancer using different registration methods suggested that thoracic vertebrae can be used for the image guidance of lung cancer images guided radiotherapy [
24]. The study of CBCT image registration in 25 patients with lung cancer found that the results of central type and peripheral type registration are different. For central type of lung cancer, protrusion registration is the best, but the spine registration is the worst. For peripheral type of lung cancer, tumor registration mark is the best, and the spine, protrusion registration is the worst [
25]. In another study, 15 cases of lung cancer patients with different anatomical region registration methods were compared by the planned CT planning tumor volume (PTV) and CBCT GTV. It was concluded that the accuracy of bone registration in the same registration range is worse than that of grayscale registration [
26]. Ottosson et al. (2010) proved that the pendulum error measured by different registration methods is different and requires different out-of-the-place boundaries. Manual registration is manually regulated by doctors to be exactly the same as the planned CT image, which is considered to be the most accurate registration method, but takes a long time and affects the work efficiency [
27]. In this study, compare to automatic registration, the results showed that there are no statistical differences in the direction of x-axis, y-axis and z-axis, regardless of the selection of the whole lung, tumor, vertebral body, or the affected lung in the manual registration. We speculate the reasons are that in the registration process, doctors also refer to different anatomical signs, and give only more comprehensive consideration. If the tumor boundary is clear, it will be first confirmed that the tumor registration is consistent. If the tumor is closely related to the endangered organ, such as the spinal cord, the vertebral body may be used as a reference sign for a larger proportion. Therefore, there are no statistical differences between manual registration and automatic registration of different parts.
Nakamura et al. (2015) demonstrated that the accuracies of correlation models derived using the shortest modeling period of 20 s are almost identical to those obtained over the longest modeling period of 40 s [
28]. In terms of registration time, previous study found that the smaller the registration range, the less information needs to be integrated, and the less time takes [
29]. The registration time for tumor group and the vertebral body group is more than 1 s, the affected lung group is more than 2 s, the whole lung group is nearly 4 s, and the artificial registration group is more than 3 min, and the differences are obvious. Therefore, it is confirmed that if the scanned image is not consistent with the planned image, it will take a long time for registration. In the study, it is also found that the smaller the registration range, the more accurate the organization registration, the less care for the outside of the registration frame. According to the median number of tumor volume 80.28 ± 6.82 cc, the patients were divided into small tumor group and large tumor group. Comparing the general conditions of the two groups, it is found that the difference of left lung volume is statistically significant. The error numerical sizes of different registration methods are also compared, and the differences are not statistically significant. However, in the study, it is suggested that the peripheral small tumors with large breathing amplitude are significantly different from other registration methods indicating that we need to increase the samples for a further study.