Background
Radiotherapy is an indispensible part of lung cancer treatment. It is estimated to be necessary in 50% of patients with small cell lung cancer (SCLC) and in over 60% of patients with non-small-cell lung cancer (NSCLC) in the course of the disease [
1]. According to several studies, local control rates of over 95% can be achieved using stereotactic body radiotherapy (SBRT) [
2,
3]. However, the treatment success hinges on an accurate target volume definition [
4]. Target delineation in the lung is especially challenging due to tumor motion caused by respiration. The extent of motion depends on tumor localization and the patient’s breathing pattern. For tumors located close to the diaphragm, amplitudes of over 2.5 cm have been measured [
5,
6]. To detect tumor motion accurately, the use of four-dimensional computed tomography (4DCT) is a reliable tool. The 4D-CT generates multiple CT-images, each representing the tumor localization and extent at a certain breathing phase. Contouring of the tumor is usually performed in every single breathing phase with subsequent definition of an internal target volume (ITV) that takes the complete cycle of movement into account. There is good evidence that the use of a 4D-CT reduces motion artifacts and makes target localization more reliable compared to the 3D-CT [
7,
8]. This results in better tumor coverage and a decrease of normal tissue irradiation during the treatment [
9]. The ITV concept is commonly used for motion management. It ensures excellent tumor coverage but exposes a larger part of healthy lung tissue to radiation. Active motions management such as breathing coordination (gating) and tracking allow smaller treatment volumes and reduction of the dose in the organs at risk (OAR) [
10,
11]. However these techniques require information on the tumor position in real time and therefore more complex technical equipment. Thus, the ITV concept remains the preferred motion management technique for many clinics. A major disadvantage of the 4D-CT is the fact that outlining gross tumor volumes (GTVs) in multiple CTs can be time-consuming, especially if a large tumor volume is contoured. Thus, since the introduction of the 4D-CT, alternative contouring methods have been discussed. On the one hand the ITV could be contoured on fewer breathing phases (usually the extreme ones) [
9]. On the other hand the ITV might be contoured in average (AIP) or maximum intensity projection (MIP) [
12,
13]. Several phantom studies concluded that MIP and AIP are reliable tools for target definition [
14,
15]. However there is a lack of clinical studies confirming these findings. A few studies based on small patient collectives (< 20 patients) showed that contouring in MIP might be an adequate option for smaller lung cancers (UICC Stage I) [
12,
13,
16]. However a study by Cai et al. [
17] using dynamic magnetic resonance imaging as reference concluded that 4D-CT MIP image might cause underdosing due to inaccurate target delineation. Thus the present literature is inconclusive and does not allow any clear conclusions. This study was performed to assess the error of MIP and AIP with special emphasis on tumor localization analyzing a large patient collective in combination with simulations of patient movements on a self-developed lung phantom [
18,
19].
Discussion
Our results show that ITVs contoured in MIP and AIP differ significantly from ITVs contoured in 10 phases of a 4D-CT. In the clinical study, average deviations of approximately − 25% were observed with even larger differences for tumors that border the mediastinum, the chest wall or the diaphragm. The data acquired with the lung phantom shows that ITV10 overestimates the target volume to a certain degree. ITVMIP and ITVAIP on the other hand underestimate the target volume und therefore do not reliably encompass the tumor tissue in all cases. Differences between ITV10 and ITVMIP/AIP were substantially larger in the clinical study compared to the phantom study.
A study of Park et al. [
14] analyzed the accuracy of MIPs for various target motions using a programmable lung phantom. Two targets inserted in a cork block were moved with irregular target motions along the superior-inferior direction and the two-dimensional target span in moving direction was measured and compared to the theoretical values. They concluded that the MIP accurately reflects the range of motions for regular target motions. However the validity of the results for the clinical practice is limited, since target motion was simulated in only one direction, no volume assessment was performed and tumors in patients seldom undergo a regular movement pattern. Simon et al. [
15] used a lung tumor phantom to simulate anterior-posterior movements to compare AIP and MIP ITVs to calculated theoretical values. The error on volume assessment ranged from.
− 40% to − 9% for the AIP and from − 3 to 12% for the MIP. The average deviations from the calculated values measured in our study in the lung phantom were also within this range (Table
1). The authors concluded that MIPs could be used for target definition of moving targets in a 4D-CT, as it seems to encompass the tumor movement. However, before this conclusion is drawn, the question must be raised whether these results from idealized phantom conditions can be transferred to the clinical situation where serrated tumor shapes, complex tumor movements and irregular density distributions occur.
The available clinical date is inconclusive and based on small groups of patients only. Bradley et al. [
12] compared MIP, AIP and helical 4D-CT images of 20 inoperable peripheral stage I lung tumors to determine the best definition method for stereotactic body radiation therapy. MIP-defined ITVs were significantly larger than helical and AIP defined ITVs. They concluded MIP is superior to AIP in order to depict tumor motion. However, since no comparison to the ITV
10 was done, the question whether the actual tumor is represented accurately by the ITV
MIP remains unclear in this study. A study by Murihead et al. [
13] collected 4D-CT data from 14 patients with NSCLC. ITVs were contoured in 10 phases of a 4D-CT and in MIP. The ITV
10 served as a reference volume to evaluate the precision of the MIP. In average 19% of the ITV
10 were not enclosed by the ITV
MIP. This is in accordance with our findings showing an insufficient coverage of the ITV
10 in MIP and AIP. The authors proposed the use of the MIP image target delineation for patients with stage I disease, since only minor deviations (6.1%) occurred in this subgroup, which consisted of 2 patients in the study. Contrary to this, in our study small targets (Ø < 2.5 cm in the clinical study, Ø 1 cm in the phantom study) resulted in the poorest conformity between MIP and the ITV
10 (Fig.
3). Underberg et al. [
16] analyzed 4D-CT data from a phantom and from 11 patients with small Stage I lung cancer. ITVs generated in all 10 phases were compared with ITVs generated in MIP. The average ratio between ITV
10 and ITV
MIP was 1.04 for the phantom study and 1.07 in scans of the patients. The center of mass differed by only 0.4 or 0.5 mm, respectively. The authors concluded that MIPs are a reliable clinical tool for generating ITVs from a 4DCT data set. Even though not explicitly mentioned by the authors it appears in the figures that narrow window settings (e.g. mediastinal window) have been used for contouring. As shown in current literature these windows do not accurately reflect moving targets and might have a major effect on the results [
18] .
According to our results MIP does not accurately depict the target volume as contoured in each of the 10 phases of a 4D-CT. The deviations between ITV
10 and ITV
MIP/AIP in the clinical study (MIP: – 20.2% AIP: -33.7%) were almost twice as large as in the phantom study (MIP –10.0% AIP -18.7%). Even though the MIP reflected the calculated values in the phantom study well, relevant underestimation of the target size needs to be expected in the clinical practice. This is especially true if the tumor borders the mediastinum, the chest wall or the diaphragm and if tumors show an extensive motion amplitude. For these tumors the deviations were particularly large. The reason for this is a loss of contrast between tumor and surrounding tissue by using maximum values for every voxel, leading to underestimation of the tumor in the overlapping areas. The large deviations using MIP for tumors bordering soft tissues could be also observer dependent, in particular as a visual approach was followed instead of automatic contouring. Since extreme movement of tumors bordering soft tissue impedes definition of an ITV, other treatment options like robotic radio surgery or breath hold techniques should be taken into consideration, in these situations [
22,
23].
The 4D-CT has been adopted as a standard modality for target delineation in lung tumors because it represents moving targets significantly better than slow 3D-CTs. Nakamuru et al. [
7] evaluated in 32 lung cancer patients the geometrical differences in target volumes between slow CT- and 4D CT- imaging for lung tumors. They observed that target volumes acquired in slow 3D-CTs are approximately 25% smaller compared to target volumes contoured in a 4D-CT. In our study MIP ITVs were on average 20.2% smaller than the ITV
10, which, by extrapolation, can be compared to the previous reported difference between the slow 3D-CTs and ITV
10. Thus, the use of MIP comes with a risk of losing additional and relevant information obtained by analyzing all phases of the 4D-CT.
The current study focuses solely on the impact of MIP and AIP on definition of the ITV. However, it needs to be also taken into account that the data set used for treatment planning has an important effect on the dose distribution within the tumor and the OAR [
20,
24]. Due to respiration-induced density variations within the ITV 3D, dose calculation based on free-breating-, MIP-, AIP- or mid-ventilation CT datasets only estimates the actual dose in the tumor [
10]. The dosimetric characteristics of plans based on AIP and mid-ventilation CTs are reported to be similar to those of FB- CTs [
25]. Treatment plans calculated on a MIP CT dataset on the other hand may not be not appropriate for OAR dose assessment [
20]. A promising approach to cope with density variations is the use of 4D-CT treatment planning with respiration-correlated assignment of the treatment plan’s monitor units to the different respiration phases of a 4D-CT and subsequent rigid and non-rigid registration [
10].
A potential limitation of this study is the impact of interobserver variability on the contouring of lung tumors in a 4D-CT. Louie et al. [
26] showed that the percentage shared internal target volume of 6 physicians contouring 10 different tumors ranged from 31.1 to 83.3%. Therefore the observed effect might differ in some cases. Nevertheless we cannot recommend the MIP (and AIP) as a standard procedure in clinical practice, since relevant underestimation of target motion and tumor volumes may occur. Whenever MIP is used for contouring, we strongly recommend to double check that the ITV encompasses the delineated target in each of the 10 phase of the 4D-CT.