Comparison of carina-based versus bony anatomy-based registration for setup verification in esophageal cancer radiotherapy

Treatment policy for esophageal cancer has substantially changed in the last decade, with radiotherapy playing an increasingly important role in both neo-adjuvant and definitive treatment [1, 2]. Esophageal radiotherapy encompasses considerable geometrical uncertainties due to setup errors, position variation of the esophageal target volume, and organ motion. Image-guided radiotherapy (IGRT) was developed to reduce these geometrical uncertainties by acquiring images of the patient’s anatomy directly prior to treatment and comparing these with the anatomy during treatment planning, by rigidly registering the 3-dimensional (3D) planning computed tomography (pCT) with the kilo−/megavoltage (kV/MV) cone-beam CT (CBCT) on the bony anatomy (i.e., vertebrae) [3]. Target volume misalignments are subsequently assessed and setup corrections are made (generally) through a couch shift.

Ideally for esophageal tumors, a tumor-based registration is used. However, with the limited soft-tissue contrast in CT, and especially CBCT with its even lower contrast resolution, the tumor cannot be discriminated reliably from its surrounding tissues. For standard clinical use, the well-visible bony anatomy (i.e., vertebrae) is currently used as a tumor surrogate in the CBCT-guided setup verification procedure. Intra/para-tumoral fiducial markers have been investigated as a tumor surrogate for setup verification. Unfortunately, a recent study at our department demonstrated that rigid marker-based registration is unfeasible due to tissue deformation [4].

Large intra- and interfractional tumor position variation is observed at the daily treatment fractions when a setup according to bony anatomy in esophageal cancer is used [4‐7]. To account for such geometrical uncertainties in the absence of adequate correction strategies, the use of generous safety margins is necessary. Nevertheless, these substantial margins expose nearby organs to increased radiation doses, which might increase toxicity and impede the use of dose escalation [2, 8, 9].

With its close proximity to the esophagus and the mediastinal lymph nodes, clear visibility on CT and CBCT, and its ability to move during respiration as opposed to the rigid and stable vertebrae, the carina might be superior to the vertebrae as a tumor surrogate for setup verification and might reduce interfractional position variation, as was demonstrated in lung tumors [10, 11].

The aim of the present study was to investigate the feasibility and assess the target coverage and geometric accuracy of a carina-based registration for CBCT-guided setup verification in esophageal cancer IGRT, compared with current practice bony anatomy-based registration. Moreover, the use of fiducial markers allowed to quantify interfractional position variation in all directions, including the otherwise hard to determine craniocaudal (CC) direction, and made it possible to compare different subgroups based on their locations in the esophagus.

This study is the first to report on the feasibility and accuracy of a carina-based registration for setup verification in esophageal cancer IGRT with the aid of fiducial markers. This was accomplished by comparing pCT-CBCT image registrations based on the bony anatomy (i.e., the current clinical standard) and on the carina. With the use of fiducial markers as a surrogate for tumor position, instead of delineated target volumes as used in previous studies, uncertainties from intra-observer delineation errors were avoided. Moreover, the use of fiducial markers allowed us to quantify the SE and sdREs in all directions, including the otherwise hard to determine CC direction [5, 17, 18]. Bony anatomy registration showed to be superior in terms of target coverage and demonstrated a smaller SE and, consequently, a smaller Σ and smaller required margins in most esophageal regions. Only for small tumors entirely confined to the mid-thoracic region, a small advantage in favor of a carina-based registration may exist.

The carina-based registration has been documented numerous times for IGRT in lung tumors [17, 19, 20]. First described in a series of 30 lung cancer patients, only including centrally located lesions, a high interobserver agreement with an excellent reproducibility was demonstrated using a carina-based registration, compared with a bony anatomy-based registration generating lower levels of reproducibility among observers [19]. However, in terms of target coverage, both registration methods provided complete coverage with no reported geographic misses, indicating that the carina or the bony anatomy could equally be used for accurate image registration in lung cancer patients [19]. Another series found a significant margin reduction in the CC direction from 1.24–0.82 cm using a carina-based registration for mediastinal lymph node irradiation in lung cancer IGRT [20]. A more recent study showed that, when a target volume consists of a lung volume and a mediastinal volume, a carina-based registration can be a superior setup verification method in terms of target coverage compared to the bony anatomy solely [17].

In contrast, for esophageal tumors, only one early study reported on carina-based registration in 20 patients [18]. This series compared different region of interest volumes to determine the optimal volume for an esophageal tumor registration; however, it suffered from several errors inherently connected with the use of a delineated target volume as a surrogate for tumor position and a cubic clipbox defined region of interest instead of a modern-shaped region of interest (i.e., a so-called mask) for registration [18]. In that series, smaller overall Σs in all directions were found; LR: 1.9 mm, CC: 2.3 mm, AP: 2.6 mm, when using a carina-based region of interest, compared to our Σs; LR: 3.3 mm, CC: 3.6 mm, and AP: 3.9 mm. This might be explained by the voluminous delineated structures, not taking deformation into account and making small differences undetectable. By calculating Pearson’s correlation coefficient, a poor correlation was found between the PTV-carina registration for lower located tumors and a good correlation for mid-thoracic tumors, which is similar to our findings (despite the fact that their findings were based on whole tumor volumes).

In the our series, an automatic carina-based registration was possible in 94.9% of registrations, in the remaining 5.1% a cumbersome manual registration was necessary making it prone to interobserver variation and human error. Further, this is considered impractical in the current clinical context. After registration, an adequate target volume coverage was reached in only 91.1% of the cases. Additionally, in two patients with distal tumors, an inadequate coverage was seen in over 1 of 3 registrations; presumably this is due to an unfavorable carina position for setup verification in esophageal cancer IGRT, because of patient-specific anatomy. This inadequate coverage might lead to a clinically relevant dosimetric impact on total dose; however, no conclusion can be drawn since this is beyond the scope of our investigation, but might be interesting for future research. However, with one registration approach showing 100% target coverage, one could argue the added value of such a comparison.

The only significant advantage of a bony anatomy-based registration was seen in the AP direction for distally located markers (Σ_carina: 4.5 mm vs. Σ_bony: 1.9 mm) (p = 0.004). This advantage might be due to the less concordant movement between esophagus and carina than initially expected, more pronounced in the distal region where a comparable rigidity in AP direction of the esophagus and the vertebrae is seen due to their similar anatomical borders. Moreover, enhancement of this discordance is imaginable due to the increased distance between the carina and lower esophagus [10, 18]. In the lower esophageal regions, gastric filling might introduce some interfractional position variation as well, this was recently investigated in a dosimetric study [21]. They stated that when an PTV margin of 1 cm is applied with a bony anatomy registration, dietary instructions do not contribute to optimal target coverage. CBCT before treatment does provide information on interfractional and interindividual variations in stomach volume for GEJ tumors but there is currently no need for adaptive treatment planning [21]. The SE and sdRE of interfractional marker position variation was expected to be affected by registration of the time-averaged CBCT with a snapshot pCT and possible artifacts (e.g., due to swallowing) or several anatomical changes during the radiation treatment course. However, since the same patients were used for both registrations, the SEs and sdREs were affected similarly, making comparison possible.

For our limited group of 24 patients, we calculated pro-forma margins, only including this interfractional position variation of the esophageal tumor (e.g., encompassing all four esophageal regions). Bony anatomy registration yielded overall margin contributions of 8.9 mm (LR), 11.9 mm (CC), and 6.8 mm (AP) (Table 2) [4]. The carina registrations yielded overall margin contributions of 10.4 mm (LR), 10.8 mm (CC), and 11.8 mm (AP), respectively. Thus, when using the carina instead of bony anatomy for patient setup verification, PTV volumes should be even larger. However, the margin estimate should be interpreted with caution due to the small sample size, inclusion of all four esophageal regions, the potential correlation between the marker position variations in one patient, and the exclusion of other errors (e.g., delineation, intrafractional) from the calculation. Only for mid-thoracic tumors was a required margin seen of only 8.0 mm vs. 8.9 mm in LR direction, 6.1 mm vs. 8.6 mm in CC direction, and 6.1 mm vs. 9.6 mm in AP direction, for a carina versus a bony registration respectively. Thus, it may be advocated to use a carina-based registration for tumors located in this mid-thoracic region. Nevertheless, target volumes in esophageal cancer typically extend across regions (including proximal and distal regions), making the net benefit of a carina registration clinically unfavorable compared with a bony anatomy registration.

The limitations of this study are the independent analysis of all markers separately, with no per patient/tumor correlation investigation being performed. A possible correlation between fiducial marker position variation and tumor/patients’ characteristics is possible, e.g., tumors in different regions, or patients with different features might show different movement properties. However, our sample size was not large enough to enable such a comparison. Another concern about carina registration is the possibility of motion artifacts resulting from the carina being captured during different positions of the respiratory cycle. This physiological motion on CBCT could cause a SE, as illustrated in a stereotactic liver study [22]. Using an averaged 4D-pCT for carina-based registration might result in smaller SE and sdRE.

Ideally, a direct daily marker-based registration would allow to account for interfractional tumor position variation. A previous study performed at our department, investigated and quantified the migration of these fiducial markers on the same data set as used in the current study [4]. They investigated the variation of pairwise distance between markers over the treatment course. Based on these results, they found some tissue deformation potentially induced by tumor regression, radiation toxicity, and/or different anatomical changes such as e.g., stomach filling over the treatment course, but with no evidence for marker migration. Nonetheless, because of this tissue deformation marker-based registration was shown to be infeasible. Even with deformable registration methods available, patient setup correction remains limited to a rigid transformation of the patient. This results in rigid bony anatomy registration still being the gold standard for routine daily clinical practice [4].

For future perspectives, the investigation of other registration volumes (e.g., PTV volume, diaphragm) and the individualization of image guidance methods on the basis of tumor location, may reduce residual error allowing the use of smaller margins. Additionally, the use of MRI guidance may allow direct tumor-based registration, reducing margins even further. At our institution, a bony anatomy-based registration with a visual marker assessment remains the optimal verification, ensuring that patient setup inaccuracies are corrected accurately and that any potential anatomical changes are identified and addressed on a day-to-day basis.

A carina-based registration in comparison with a bony anatomy-based registration for esophageal cancer IGRT, results in inadequate target coverage in 8.9% of cases. Furthermore, large SEs and sdREs resulting in larger required anisotropic margins were seen after a carina-based registration. The carina-based registration might be slightly favorable only for tumors entirely confined to the mid-thoracic region. Therefore, our data endorse the use of bony-anatomy based registration over a carina-based registration for routine setup verification for esophageal tumor IGRT.