Primary and metastatic hepatic malignancies are commonly treated by surgery, but radiation therapy is also one of options as non-surgical modalities. It has been demonstrated that radiation therapy is feasible and the outcomes are promising [1, 2]. However, due to respiration liver motion up to 3 cm [3] is one of obstacles to accurately localize the target. Moreover, respiratory-induced tumor motion is known to be anisotropic, thus individual determination of internal margin around gross tumor volume (GTV) is crucial to form an internal target volume (ITV), which can avoid both inadequate tumor coverage and unnecessary liver parenchymal irradiation for individual patient. Four-dimensional computed tomography (4D CT) is one of appropriate approaches to estimate and determine ITV for tumor with respiratory motion [4, 5].

4D CT has been widely used in lung cancer to determine ITV [6, 7]. Ideally, ITV should be delineated by manually contouring GTV in all 10 breath phases of a 4D scan image sets to form ITV, which is the most accurate tool to determine ITV, but it is a time-consuming and labor-intensive task. To reduce the workload of contouring multiple GTVs, one solution is to contour only two extreme phases at end-inhalation and end-exhalation and then to sum of the two becoming ITV [8, 9]; and the other is to use the post-processing tools of maximum intensity projection (MIP) and minimum intensity projection (MinIP) from 4D CT data sets to generate ITV. MIP-based ITV delineation is performed on a single 3-D CT data set, where each pixel in this set represents the brightest object encountered by corresponding voxels in all volumetric 4D CT data sets, for instance, MIP-based ITV delineation for lung cancer, which was recommended as a reliable tool and a good first estimation [10–12]. Conversely, MinIP-based ITV is on the CT set, where each pixel represents the lowest data value in the volumetric data [10].

MIP and MinIP methods seem not suitable for liver cancer because most tumors in the liver have similar attenuation to the normal liver parenchyma and therefore are not easily discernable. Contrast-enhanced CT scan has been routinely used for radiation oncologists to differentiate the tumor from normal tissues. It should be noted that regardless of with contrast enhancement or not, most liver tumors present inhomogeneous density, either because of the inherent nature of the tumor, such as the routes of blood supply, vascular volume and permeability, or because of areas of fluid, hemorrhage, and necrosis within tumors, or because of secondary change due to treatments, for example, iodine deposition as a result of transcatheter arterial chemoembolization (TACE). Recently, Beddar et al described a simple method for 4D CT acquisition by using synchronized intravenous contrast injection to improve the accuracy of liver tumor delineation. By this way most liver metastases and cholangiocarcinomas can be identified on image of portal venous phase, while HCC, most visible in the delayed phase [13].

In theory, for tumors with homogeneous hyperdensity or hypodensity comparing to the surrounding normal liver, MIP or MinIP projections should accordingly reflect the tumor trajectory across all time-resolved data sets. Visualization of tumor with mixed-density means the tumor border should be discernable; regardless which part of it is more hyperdense or hypodense than the adjacent liver parenchyma. Using MIP technique or MinIP technique only definitely misses the spatial information of the moving liver tumor. Therefore, our hypothesis is to combine MIP and MinIP, which may fit for the situation of mixed tumor density. Thus, the purpose of this study is to evaluate the feasibility and accuracy of the MIP and/or MinIP-based ITV delineation in 4DCT scans for liver tumors.

MIP method as an image post-processing is based on more complex algorithms and can be used for generating three-dimensional vascular reconstructions [14, 15]. In lung cancer, MIPs are believed to be a reliable tool to generate ITVs from 4D CT data sets [10], however, it is mandatory to modify each individual MIP to improve ITV delineation for tumors adjacent to the thoracic wall, mediastinum, heart, or diaphragm [11, 12]. In the current study, we also had to verify ITVs contoured on MIP and MinIP CT by overlaying it on a movie loop displaying 4D CT data and then editing it, especially for those closed to adjacent organs. Mancosu had recently proposed a semiautomatic technique, which allowed for inclusion of the residual part of ITV covered by liver and spleen cupola when using MIP algorithm. It was validated on phantom and selected patients, which revealed this possibility when lesion located near liver and spleen cupola by performing only the contours on MIP series [16]. Thus, the dedicated software needs to be developed to exclude diaphragm and chest wall in some breathing phases using 4D CT for better tumor MIP/MinIP images.

Theoretically, for tumors with homogeneous hyperdensity or hypodensity compared to the surrounding normal liver, MIP or MinIP projections should accordingly reflect the tumor trajectory across all time-resolved data sets. In patients when the lesions were homogenous low CT density, located intrahepatic, not adjacent to perihepatic organs and also small size (5^th, 6^th, 8^th and 12^th patient), using MinIP was also an appropriate tool with good MI of 0.92 ± 0.04 and low proportion of V_under of 0.08 ± 0.04. Thus, ITV_MinIP method can be used for fast contouring of liver tumor with homogeneous low CT density including respiratory motion.

However, a number of liver cancers present inhomogeneous density, results from this study showed combined MIP and MinIP fit excellently in this situation. In current study we compared MIs and proportions of V_under resulting from ITV_2Phase, ITV_MinIP, ITV_MIP and ITV_2M contoured by 4 approaches, and found that the closest to ITV_AllPhases was the combined one (ITV_2M) of ITV_MinIP and ITV_MIP, which were contoured on MinIP and MIP series of 4D CT set because it resulted in the highest MI and lowest proportion of V_under. Moreover, the size of tumor and the ratio of tumor vertical diameter over cranio-caudal movement did not have influence on MI or proportion of V_under when using ITV_2M. In other words, no matter how big the tumor was, and the tumor vertical diameter over cranio-caudal movement was small or big, the tool of ITV_2M could always result in the best outcome.

For moving target it was also a practice to sum 2 GTVs to generate ITV, one contoured on CT image set acquired after end-inhale and holding breath and the other contoured on CT acquired after end-exhale and holding breath when 4D CT was not available. The similar method was also used for 4D CT, i.e., to contour only two extreme phases at end-inhalation and end-exhalation. However, possible hysteresis effects would be neglected as occurred in lung cancer [17, 18]. Seppenwoolde and colleagues [18] reported that when the tumor was small and had a large range of motion, the separation between the positions of the images of inspiration and expiration phases was relatively obvious and the information of the intermediate breathing might not be comprehensive. Besides, the combined images of the two time phases might omit the lag of the tumor. The phenomenon was caused by the time difference among the document recorded by the computer, the transition of respiratory cycle and the transition between inspiration and expiration in a respiratory cycle. Xi [9] reported the feasibility of using limited 4D CT images for treatment planning for liver radiotherapy. As recognized by the authors, deriving ITV by two extreme phases was reasonably safe only for low and medium tumor motion amplitude (< 1.6 cm). The tumor motion in cranio-caudal direction between Xi's study and ours' were comparable, but the tumor size of our data (152.20 cm³ ± 242.85 cm³) was more diverse than Xi's (70.36 cm³ ± 66.23 cm³). Xi did not investigate the influence of tumor size on ITV determination using 4D CT data. Our study did find a smaller volume of ITV_2Phase (186.8 cm³) than that of ITV_AllPhases (200.8 cm³) with significant difference (p = 0.004). And also we found that MI and proportion of V_under were influenced by the tumor volume and the ratio of tumor size over tumor motion magnitude significantly. Whereas, ITV_2M was the closest to ITV_AllPhases, and MI and proportion of V_under were not influenced by tumor volume and the ratio of tumor size over tumor motion magnitude when using tool of ITV_2M.

To reduce the workload of contouring multiple GTVs in 4D CT data sets, contouring only two extreme phases is appropriate only when tumor volume is big and tumor motion magnitude is relatively small. For hepatic malignancies with inhomogeneous density we found that the method of using ITV_2M was a more reliable and appropriate tool for generating ITVs from 4D CT data sets, compared to the others.