The spine and carina as a surrogate for target registration in cone-beam CT imaging verification in locally advanced lung cancer radiotherapy

Recent advances in systemic and radiotherapy treatment have resulted in improved survival for patients with inoperable locally advanced lung cancer.¹ But still, nearly half of the patients will experience locoregional relapse.² The accuracy of different steps in radiotherapy treatment preparation and execution have a strong impact on local control.³ With the introductionof computed tomography (CT), positron emission computed tomography (PET CT) and four-dimensional (4D) CT simulation, the target delineation accuracy increased.⁴ Modern radiation techniques have enabled a more conformal dose delivery, the dose to normal tissue was reduced and the radical treatment of more advanced N3 disease and/or dose escalation has become possible.⁵ Because of the steeper dose gradient and smaller safety margins, the accuracy of treatment delivery became increasingly important. For daily treatment position verification, cone beam CT (CBCT) largely replaced electronic portal imaging because of better soft-tissue visibility. The alignment of the treatment image with the planning CT (pCT) is usually performed by radiation therapists (RTTs), who are not trained in target determination. The fast interpretation of CBCT is also challenging due to the lower quality of CBCT images (no i.v. contrast) and changes with or adjacent to the tumour during treatment.⁶ Manual tumour matching is subjected to strong inter-observer variability even among radiation oncologists.⁷ The optimal surrogate structure for image matching has not yet been established. Spine alignment is feasible and reproducible, but shows poor correlation with tumour position.⁸ Recently, the carina surrogate alignment was explored and showed superior reproducibility compared to spine alignment.⁷

We performed a retrospective study to determine the accuracy of the carina vs. spine registration compared to target (primary tumour and lymph nodes) registration as a reference. To avoid interobserver variability, reference registration was performed by individual thoracic radiation oncologist. To consider tumour and normal structure variations during the course of treatment and their possible impact on image registration, the 1^st, 10^th, 15^th, and 20^th fraction CBCT were included in the registration analysis.

Since the introduction of CBCT into treatment verification, new insights to target position uncertainties have evolved⁶, suggesting that image guidance could currently be the weakest link in the radiotherapy procedure.^3,9 Only a few studies have explored the carina as a surrogate in volumetric lung image guidance^7,10,11 and there is a lack of knowledge in this field. Here we present new data on the accuracy of the carina vs. spine surrogate compared to the target alignment as a reference in image registration.

A separate analysis of the spine and carina registration set-up differences compared to the reference (target) registration showed the smallest differences in the VRT and LAT directions, with the largest deviation in the LONG direction for both strategies (Fig. 2). The differences were not significantly different in time and no time trend was detected for the cohort, so we evaluated all the measurements together and again showed a good registration match in the VRT and LAT directions, but in LONG direction, the range of misalignment reached above 5 mm in both registration strategies. The greatest area of uncertainty in the LONG direction was also shown in the study by Ottosson et al. where they compared the spine and target registration in free-breathing and breath-hold CBCT in locally advanced lung cancer patients. The largest intra- and inter-fractional misalignments were found in the LONG direction, independent of the registration method.¹² Although we found a smaller mean LONG misalignment in the carina vs. spine registration (1.78 vs. 2.44 mm), uncertainties in image registration should be considered for both strategies, with the enlargement of the PTV in the craniocaudal (CC) direction.

To determine the best registration match compared to the reference, we compared differences in the set-up deviations for the spine and carina alignment. Both strategies were equally accurate in the VRT direction, but the carina was more accurate in the LONG and LAT directions in all measurements. The accuracy of the spine vs. carina registration was also tested in a study by a Canadian group, where four independent observers automatically and manually aligned the first fraction CBCT with the pCT in 30 lung cancer patients.⁷ They used spine, carina and tumour registration strategies. Automatic spine and carina registrations provided similar tumour coverage, with the tumour inside the ITV in 60% of observations. The same group, in a second study, verified the tumour (T) and lymph node (LN) coverage following spine and carina registration for initial, middle, and final fraction CBCT. Both strategies improved the combined target coverage throughout the treatment course compared to tattoo alignment. Carina better improved the combined coverage and showed significantly superior nodal coverage compared to the spine, without compromising primary tumour coverage.¹⁰ With a significantly better registration match in the LONG and LAT directions, our data supports the suggestion from the Canadian group that the carina may be superior to the spine in the image guidance of locally advanced lung cancer.

The second Canadian study showed similar primary tumour coverage, regardless of the registration strategy.¹⁰ But in our study, the registration accuracy was influenced by tumour location. Both the spine and carina alignment showed smaller set-up differences for centrally located tumours with the difference being more significant in the carina registration LONG (66% fractions) and LAT (50% fractions). For peripheral tumours, we found no difference in the accuracy of spine vs. carina alignment, but for centrally located tumours, the carina was more accurate in the LAT and LONG directions. Our data suggests that the carina is a better surrogate for centrally located tumours. Importantly, we also showed that the carina is as accurate as the spine for registration in peripherally located tumours and can be proposed as a registration surrogate regardless of the tumour location.

According to our data, carina matching can also be used regardless of the nodal stage. We found no differences in the spine, but a better match for carina alignment in N2 and N3 disease. In advanced nodal disease, the carina showed superior registration vs. the spine in LAT and LONG. The finding is in concordance with the second Canadian study, where carina matching improved the node coverage compared to spine registration.¹⁰ Importantly, we found no set-up difference for N0 disease suggesting that the carina and spine can equally be used as a surrogate in this stage. As there was only a single case of early nodal disease, we cannot make any conclusions for N1. The reliability of spine matching for LN coverage was investigated in the study by Mohammed et al., where an equal geographical target miss was found for spine vs. combined target matching, but inferior LN coverage in the case of tumour matching, based on a weekly 4D CBCT registration.⁹ Because the same geographical miss was shown for hilar and mediastinal lymph nodes, we can hypothesize that carina matching could safely be used in N1 as well.

Target alignment should be “the ground truth” and have an impact on the therapeutic ratio,^3,12 but is rather difficult to implement clinically. Due to the poor soft tissue contrast on CBCT and tumour changes during the course of treatment, only half of the tumours can clearly be contoured using CBCT.¹³ For these reasons, we used contoured-based registration for reference. A similar approach was used for target coverage and geographical miss assessments studies^9,10 and target registration in the study by Ottosson et al.¹² Direct tumour registration is also impractical because it is time-consuming and unreliable due to high intra-observer variability.⁷ In contrast, the carina is clearly visible on the CBCT and automatic carina alignment shows high reproducibility, superior even to spine alignment.⁷ Although the carina shows some respiratory movement, especially in the CC direction, it is still an excellent surrogate for directly adjacent mediastinal lymph nodes. Preliminary data also suggests a good correlation between the carina and GTV motion.^7,11

We acknowledge the limitations of our study, which was retrospective and relatively small. We also present a heterogeneous group of patients, though they represent real everyday radiotherapists’ lung cancer patients. In our study, CBCT image changes were not systematically determined. Because the majority of changes appear in the first five weeks of radiotherapy,¹⁴ we analysed CBCT images from the 1^st, 10^th, 15^th and 20^th fractions. Maybe the 25^th and 30^th fraction CBCT should have been included since we found cases for plan modification in the 6^th week of treatment. Nevertheless, we detected no significant set-up differences in time. The rate of plan adaptation was low (7.5%), suggesting the need for a “traffic light” protocol implementation.⁷

In conclusion, carina registration was shown to be feasible, fast, and reproducible. Our data shows that, compared to target registration, carina is equally as accurate as spine registration, with superior accuracy in the LONG and LAT directions. The carina is a better surrogate than the spine for alignment in centrally located tumours, N2 and N3 disease. Although spine alignment can equally be used in N0 and peripheral tumours, for simplicity we propose that the carina should be the primary surrogate for the target in image guidance in locally advanced lung cancer. The next step should be to incorporate carina registration uncertainties into the PTV margin.