Abstract
Intensity-based 2D/3D registration using kilo-voltage (kV) and mega-voltage (MV) on-board imaging is a promising approach for real-time tumor motion tracking. So far, the performance of the kV images as well as kV-MV image pairs for 2D/3D registration using only one gantry angle (in anterior-posterior (AP) direction) has been investigated on patient data. In stereotactic body radiation therapy (SBRT), however, various gantry angles are typically used. This study attempts to answer the question of whether automatic 2D/3D registration is possible using kV images as well as kV-MV image pairs for gantry angles other than the AP direction. We also investigated the effect of additional portal MV images paired with kV images to improve 2D/3D registration in extracting cranio-caudal (CC) and AP displacement at arbitrary gantry angles and different fractions. The kV and MV image sequences as well as 3D volume data from five patients suffering from non-small cell lung cancer undergoing SBRT were used. Diaphragm motion served as the reference signal. The CC and AP displacements resulting from the registration results were compared with the corresponding reference motion signal. Pearson correlation coefficients (R value) was used to calculate the similarity measure between reference signal and the extracted displacements resulting from the registration. Signals we found that using 2D/3D registration tumor motion in 5 degrees of freedom (DOF) with kV images and in 6 degrees of freedom with kV-MV image pairs can be extracted for most gantry angles in all patients. Furthermore, our results have shown that the use of kV-MV image pairs increases the overall chance of tumor visibility and therefore leads to more successful extraction of CC as well as AP displacements for almost all gantry angles in all patients. We observed an improvement in registration of at least 0.29% more gantry angle for all patients when we used kV-MV images compared to kV images alone. In addition, an improvement in the R-value was observed in up to 16 fractions in various patients.
Keywords: Image guided radiotherapy, 2D/3D image registration, Intra-fractional motion, Tumor monitoring, kV-MV images
1. Introduction
Intra-fractional movement during radiotherapy sessions is considered as one of the main sources of uncertainty in dose delivery. Continuous irradiation during free breathing based on the internal target volume (ITV) is one of the commonly used methods, which makes it possible to take into account the target motion caused by breathing during irradiation of the target volume. ITV irradiation is currently the most common method for stereotactic body radiation therapy (SBRT) treatment of non-small cell lung cancer (NSCLC). Several intrafraction motion management strategies were already implemented including the correlation of internal motion with surface landmarks [1], [2], kV fluoroscopy, MR tracking and the correlation of lung motion models with surface motion [3], [4], [5].
One quantitative approach is to monitor tumor motion in real time using fluoroscopic images captured by the imaging unit mounted on the linac. The use of image-guidance techniques to reduce radiation at target volume margins may be critical for some patients in order to spare more of the surrounding healthy lung tissue. Real-time 2D/3D registration with on-board kilo-voltage (kV) images has been introduced as a promising approach to successfully track the tumor motion and can lead to sparing more healthy tissue [6], [7], [8], [9]. Künzler et al. [6] compared digitally reconstructed radiographs (DRRs) of lung tumors with MV portal images in order to obtain the displacement of the tumor during the SBRT lung treatments using a rigid intensity-based 2D/3D registration approach. Since MV images provide limited contrast for soft tissue, registration was mainly based on the visibility of the bony structures. This resulted the registration process to be challenging as tumor motion is mainly correlated to the surrounding soft structures.
In a study [10], accurate motion tracking was achieved using kV portal images in five degrees-of-freedom (DOF) including three rotational parameters, ωx, ωy, ωz and two translational parameters tx, ty with a fix tz, the translation along the imaging beam axis. The limitation of this approach was that the motion could not be resolved along the anterior-posterior (AP) direction. In another study [11], the same group tried to overcome this limitation by additionally registering the MV images acquired with the treatment beam. Using this approach, the employed electronic portal images (EPI) paired with the kV images could resolve all 6 DOF (tx, ty, tz, ωx, ωy, ωz) using NSCLC patient dataset. Their results demonstrated that the use of the pair KV-MV images increase the registration accuracy in the AP direction enabling real-time, accurate, 6 DOF tumor tracking. Another research group [12] proposed a 2D/3D registration framework using a statistical model of the intensity values using two imaging modalities (kV and MV) for external beam radiation therapy of pelvic sites in a phantom study. They showed that the registration of a kV and MV portal images in AP direction was successful in the phantom experiments as well as patient images. In another study [13] the systematic errors in 6 dimensions were evaluated by comparing automatic 2D/3D image registration under various conditions encountered in clinical applications and different treatment sites including intracranial and extracranial. The performance of the image registration using kV–kV, MV–kV, and MV–MV image pairs were evaluated. They reported that the 2D/3D image registration tool which is available on the Varian Edge radiosurgery device can provide an accurate target positioning in image-guided stereotactic radiosurgery procedures. Another research group investigated 3D/3D and 2D/3D registration in order to perform six-dimensional (6D) registrations for image-guided radiotherapy [14]. They used fiducial markers placed on the surface of an anthropomorphic head and a body phantom and acquired paired kV-MV planar X-rays. 3D/3D registration achieved the best image registration accuracy among all the employed methods. The 2D/3D registration method was able to achieve slightly lower accuracy compared to the 3D/3D method, but the performance was still acceptable to be used as a treatment position verification device.
All aforementioned studies [11], [12], [13], [14] investigated the performance of the kV-MV image pairs in solving 6 DOF for 2D/3D registration using only one fixed gantry angle. However, multiple gantry angles are used to perform SBRT lung treatments with conformal techniques. Therefore, one of the challenges here is to perform 2D/3D registration for arbitrary gantry angles used in SBRT. Overlapping structures might limit the intensity-based 2D/3D registration approach at certain projection angles. The tumor and its surrounding structures are not equally visible in all directions and therefore registration becomes a challenging problem. Thus, the question of whether automatic 2D/3D registration can resolve 6 DOF using pair KV-MV portal imaging for arbitrary gantry angles has not yet been answered.
Real-time tracking techniques aim to reduce margins and spare more healthy tissue. In this study, the general goal is to use intensity-based 2D/3D registration for tumor motion tracking, which could replace the need to define larger geometric margins for the target, facilitating the transition from ITV to potentially tracking of GTV using on-board imagers. In particular, we investigated the possibility of resolving tumor motion using intensity-based 2D/3D registration using kV images as well as kV-MV image pairs at arbitrary gantry angles based on a clinical lung SBRT treatment data. In addition, the impact of adding MV image dataset to kV images in improving 2D/3D registration performance was comprehensively investigated for various gantry angles and fractions.
2. Materials and Methods
2.1. Patient Data
Data from five patients suffering from NSCLC were analyzed respectively for this study. The patient dataset included a wide range of gantry angles and different tumor positions. The patients underwent SBRT lung treatment using three-dimensional conformal radiotherapy 3D conformal radiotherapy (3DCRT) technique. The treatment was performed in three fractions with seven gantry angles irradiated per fraction at the Department of Radiation Oncology, Medical University of Vienna, Austria. A planning CT with delineated structures and daily cone beam CT (CBCT) scans were acquired for each patient. The planning CT images were obtained by a Siemens Somatom Plus 4 Volume Zoom (Siemens AG, Erlangen, Germany) at 120 kVp and 156 mAs with a pixel spacing of 0.97 mm and 4 mm slice thickness, whereas the CBCT images were obtained with an Elekta Synergy linac (Elekta Oncology Systems Ltd, Crawley, UK) equipped with an electronic portal imaging device (EPID) and an XVI system for kV imaging and CBCT acquisition. In addition, kV and MV images were acquired in order to capture intra-fraction motion.
The MV images were captured during treatment. These images represent the attenuation of the treated beam with no extra dose to the patient. 3D conformal technique using forward planning was used for the patient treatment. A fixed 90-degree angle exists between the kV and MV beam sources. The kV and MV detector panels were located 536 mm and 570 mm far from the axis of rotation, respectively. The kV image acquisition had to start manually. The kV and MV images were acquired during the treatment session at different frame rates. kV images (up to 200 images) were captured with a frame rate of 5.4 Hz and MV images were captured with a frame rate of 2.1 Hz during the beam-on status, which was around 30 to 85 seconds.
The structures such as planning target volume (PTV), clinical target volume (CTV), ITV, and body contours were delineated by a physician from the planning CT. The PTV delineation later was used as the region of interest (ROI) on kV and MV images. In addition, the planning CT and CBCT volumes were interpolated to an isotropic voxel size of 1.0 mm3 and were aligned with a 3D/3D registration using AnalyzeAVW 11.0 (Biomedical Imaging Resource, Mayo Clinic, Rochester/MN). DRR images were generated from the registered CBCT volume and were then 2D registered to the kV as well as kV-MV image pairs. The calculated translation matrix from this registration was then used to project the PTV on the kV images. Projections related to PTV masks on kV images for all five patients at gantry angles closest to AP are presented in Fig. 1 (projection angles are based on the available treated gantry angles in the first fraction of treatment for each patient). The MV images for all treated gantry angles for the first treatment fraction of each patient are shown in Fig. 2.
Figure 1.
Illustration of projections together with corresponding PTV masks on kV images for all five patients at the gantry angle closest to AP view. P. and G.A. stand for patient number and gantry angle, respectively.
Figure 2.
MV images related to all patients at various gantry angles (first fraction). The number below images represents the gantry angle in degree.
2.2. 2D/3D registration
The FIRE toolbox, an in-house developed software suite used for research in IGRT [15], has been used in this study for 2D/3D registration purpose [11] in order to find the volumetric transformation T= (tx, ty, tz, ωx, ωy, ωz), with t representing translation and ω representing rotation. The ray-casting algorithm was used for digitally reconstructed radiograph (DRR) generation. The PTV delineation was also used as the ROI on DRRs which were then used for the 2D/3D registration. The kV and MV images were interpolated with an isotropic pixel size of 1mm in order to have the same dimension as the generated DRR for the registration step and digitally reconstructed radiograph (DRR) images which were then used for the 2D/3D registration. Cross-correlation was mainly chosen as the merit function, and NEUWOA was selected as the optimizer for the optimization [16]. Some fixed values were set according to the geometry of the treatment facility: tz was set to 536 mm, which represented the distance of the detector panel of the kV system to the axis of rotation, ωx = 90°, and ωy as the actual gantry angle and ωz= 180°. For kV images, tx, ty, tz represented the ML (medio-lateral), CC (cranio-caudal) and AP directions, respectively.
All implementations were performed on a personal computer (Intel Core i5 4 core CPU of 3.4 GHz). For DRR generation, a ray-casting algorithm was implemented on CUDA on an NVIDIA GeForce GTX 780 graphics card with 8 GB RAM. In order to resolve motion in the AP axis, the kV-MV image pairs were used simultaneously in the registration procedure, and a combined merit value was optimized when comparing the kV and MV images to their corresponding DRR pair.
2.3. Motion extraction and kV-MV image pairs synchronization
Edge detection and Hough transform were used to detect circles for extracting the diaphragm motion in CC and ML direction from kV images as in [11]. The diaphragm motion was used as our reference motion signal for the evaluation of registration results. The motion annotation for MV images was done as in a previous study by H. Furtado, et al. [11], which could detect displacement in the x and y-axis of the image corresponding to the AP and the CC directions of patient. The extracted diaphragm motion and MV motion signals did not only serve as reference signals for the evaluation of the registration results, but also helped to verify the time correlation between the kV and MV image sequences. The kV and MV image sequences were acquired at different frame rates (Section 2.1). In addition, the kV acquisition was started manually which in turn could result in unknown acquisition delays between two image sets. In order to create synchronized image pairs, kV-MV images needed to be correlated in time using previously extracted motion. The delay was calculated by finding the maximum point of the cross-correlation function between the signals [11]. Then the kV images were paired with the MV-images with the closest time interval. Fig. 3 shows one example of the kV-MV image pairs (patient 2) at all gantry angles treated in one fraction.
Figure 3.
kV-MV image pairs of patient 2 related to all the gantry angles covered in the second fraction of the treatment.
We should note that the mentioned semi-manual step for temporal correlation of the signals using the cross-correlation function is required, since the linac used in this study does not have the capability to perform this synchronization automatically. However, it has been shown that there is a possibility to perform this synchronization step automatically [17]. Therefore, automatic synchronization on the modern linacs seems feasible.
2.4. Evaluation of the registration result
Motion signals were extracted from the 2D/3D registration for each gantry angle and fraction individually for the following scenarios: 1) using only kV images to extract CC displacement in 5 DOF, 2) using kV-MV image pairs to extract CC displacement in 6 DOF, 3) using the sequences of kV-MV image pairs to extract AP displacement in 6 DOF. All these assessments were performed separately for each gantry angle in each fraction. To evaluate registration results, the motion signals obtained by registration were visually compared with the annotated diaphragm and the tumor motion in CC and AP directions. Registration results were plotted with respect to their reference motion signals. Firstly, the CC displacement extracted from the registration results in 5 DOF were plotted with respect to the reference diaphragm motion when only kV images were used. In addition, when using kV-MV image pairs, the registration results obtained in 6 DOF were plotted for CC and AP displacement compared to their corresponding reference motion signals.
In this study, we investigated the similarity between two signals, including the reference motion signal and the extracted translation resulting from the registration. Pearson correlation was used to calculate the similarity measure between these two signals, as it is a very common technique for comparing alignment between signals [18]. The implementation was done using MATLAB toolbox, version R2018a, function corcoeff. For a meaningful comparison of the correlation coefficient, the same number of x-ray images were chosen from the kV and kV-MV image pairs sets. Both extracted and reference motion signals were initially re-scaled by subtraction of their mean value for better visualization. A workflow flowchart of the proposed approach can be found in Fig. 4.
Figure 4.
A workflow flowchart of the 2D/3D registration approach.
3. Results
Fig. 5 shows an example of displacements for patient 3 for various gantry angles (data from patient 3 included only 1 treatment fraction with 3 covered gantry angles). The extracted displacements related to other patients at different treatment fractions and angles can be found in the Appendix. In addition, an example of the extracted translations and rotations for CC, AP, ML directions using kV images as well as kV-MV images are presented in Fig. 6 (patient 4, fraction 3, at gantry angle 98°). As we seen in Fig. 6, extraction of AP displacement (Tz) was only possible when using kV-MV image pairs. Pearson correlation coefficients (R value) for each gantry angle were calculated for all patients and all fractions (Table 1, Table 2, Table 3, Table 4, Table 5). In order to interpret the results, first it was checked if the calculated R-value confirms the visual agreement between the reference motion and registration displacement plot. We observed a good visual agreement between the registration result and the reference signal when having R-values above 0.5 with P-values < 0.001 (highlighted values in Table 1, Table 2, Table 3, Table 4, Table 5). Therefore, the R value higher than 0.5 with P-values < 0.001 was considered significant in this study.
Figure 5.
Extracted displacement from the registration results with respect to the corresponding reference signals for patient 3, fraction 1 at gantry angles 110°, 140°, and 170°. The left column shows registration results for CC displacement using kV image set (green) vs. diaphragm motion (black). The middle column shows the registration results for the CC displacement using kV-MV image pairs (green) compared to the extracted tumor motion in CC axis (black). The right column shows registration results for AP displacement using kV-MV image pairs (green) compared to the extracted tumor motion in AP axis (black).
Figure 6.
Extracted displacements and rotations from the registration results related patient 4, fraction 3 at gantry angle 98°, (a) and (b) show translational and rotational parameters using kV images. (c) and (d) show translational and rotational parameters using kV-MV images. Translational parameters are shown in solid line and rotational parameters in dashed lines.
Table 1.
The registration results related to the Patient 1 at different gantry angles: the R-value related to the CC displacement using only kV images as well as kV-MV image pairs, and R-value related to the AP displacement when using kV-MV image pairs. The image sequences for some gantry angles were not available which is specified with x in the table. Cases with R-values higher than 0.5 included P-values < 0.001 which are shown in Bold format. P., G. A and Fr. stand for patient number, gantry angle and fraction, respectively.
|
P.1, G.A. R-value | ||||||||
|---|---|---|---|---|---|---|---|---|
| Fr. | 25° | 58° | 98° | 130° | 160° | 190° | 220° | |
| kV (CC) | Fr.1 | 0.18 | 0.56 | 0.78 | 0.91 | x | 0.64 | −0.09 |
| Fr.2 | 0.53 | 0.14 | 0.57 | 0.95 | 0.40 | 0.92 | 0.79 | |
| Fr.3 | 0.82 | −0.01 | 0.43 | 0.81 | −0.12 | 0.92 | 0.77 | |
| kV-MV (CC) | Fr.1 | 0.65 | 0.77 | 0.68 | 0.80 | x | 0.67 | 0.47 |
| Fr.2 | 0.78 | 0.87 | 0.45 | 0.64 | 0.41 | 0.71 | 0.72 | |
| Fr.3 | 0.84 | 0.84 | 0.89 | 0.89 | 0.44 | 0.74 | 0.62 | |
| kV-MV (AP) | Fr.1 | 0.52 | 0.79 | 0.57 | 0.08 | x | 0.65 | 0.77 |
| Fr.2 | 0.50 | 0.81 | 0.19 | 0.31 | 0.46 | 0.21 | 0.67 | |
| Fr.3 | 0.54 | 0.79 | 0.92 | 0.64 | 0.46 | 0.42 | 0.76 | |
Table 2.
The registration results related to the Patient 2 at different gantry angles: the R-value related to the CC displacement using only kV images as well as kV-MV image pairs, and R-value related to the AP displacement when using kV-MV image pairs. The image sequences for some gantry angles were not available which is specified with x in the table. Cases with R-values higher than 0.5 included P-values < 0.001 which are shown in Bold format. P., G. A and Fr. stand for patient number, gantry angle and fraction, respectively.
|
P.2, G.A. R-value | ||||||||
|---|---|---|---|---|---|---|---|---|
| Fr. | 12° | 40° | 80° | 120° | 150° | 180° | 205° | |
| kV (CC) | Fr.1 | 0.67 | 0.95 | 0.70 | 0.58 | 0.83 | 0.70 | 0.66 |
| Fr.2 | 0.41 | 0.57 | −0.27 | 0.41 | 0.70 | 0.70 | −0.35 | |
| Fr.3 | 0.84 | 0.87 | 0.44 | 0.67 | x | 0.80 | 0.91 | |
| kV-MV (CC) | Fr.1 | 0.78 | 0,85 | 0.72 | 0.41 | 0.77 | 0.87 | 0.75 |
| Fr.2 | 0.56 | 0,74 | 0.13 | 0.44 | 0.82 | 0.80 | 0.59 | |
| Fr.3 | 0.74 | 0.83 | 0.68 | 0.59 | x | 0.86 | 0.37 | |
| kV-MV (AP) | Fr.1 | 0.17 | 0.69 | 0.85 | 0.58 | 0.50 | 0.83 | 0.46 |
| Fr.2 | 0.24 | 0.92 | 0.92 | 0.71 | 0.62 | 0.81 | 0.81 | |
| Fr.3 | 0.66 | 0.86 | 0.86 | 0.80 | x | 0.84 | 0.66 | |
Table 3.
The registration results related to the Patient 3 at different gantry angles: the R-value related to the CC displacement using only kV images as well as kV-MV image pairs, and R-value related to the AP displacement when using kV-MV image pairs. The image sequences for some gantry angles were not available which is specified with x in the table. Cases with R-values higher than 0.5 included P-values < 0.001 which are shown in Bold format. P., G. A and Fr. stand for patient number, gantry angle and fraction, respectively.
|
P.3, G.A. R-value | ||||
|---|---|---|---|---|
| Fr. | 110° | 140° | 170° | |
| kV (CC) | Fr.1 | 0.97 | 0.97 | 0.98 |
| kV-MV (CC) | Fr.1 | 0.97 | 0.98 | 0.94 |
| kV-MV (AP) | Fr.1 | 0.83 | 0.41 | −0.71 |
Table 4.
The registration results related to the Patient 4 at different gantry angles: the R-value related to the CC displacement using only kV images as well as kV-MV image pairs, and R-value related to the AP displacement when using kV-MV image pairs. The image sequences for some gantry angles were not available which is specified with x in the table. Cases with R-values higher than 0.5 included P-values < 0.001 which are shown in Bold format. P., G. A and Fr. stand for patient number, gantry angle and fraction, respectively.
|
P.4, G.A. R-value | ||||||||
|---|---|---|---|---|---|---|---|---|
| Fr. | 35° | 80° | 110° | 140° | 170° | 200° | 230° | |
| kV (CC) | Fr.1 | 0.79 | 0.59 | 0.58 | 0.98 | 0.40 | 0.86 | 0.94 |
| Fr.2 | 0.55 | −0.20 | 0.56 | 0.90 | 0.75 | x | x | |
| Fr.3 | 0.70 | 0.35 | 0.63 | 0.82 | 0.79 | 0.83 | 0.85 | |
| kV-MV (CC) | Fr.1 | 0.72 | 0.57 | 0.93 | 0.89 | 0.69 | 0.60 | 0.82 |
| Fr.2 | 0.62 | 0.81 | 0.97 | 0.94 | 0.72 | x | x | |
| Fr.3 | 0.74 | 0.82 | 0.88 | 0.88 | 0.79 | 0.60 | 0.66 | |
| kV-MV (AP) | Fr.1 | −0.07 | 0.52 | 0.69 | 0.53 | 0.62 | 0.27 | 0.47 |
| Fr.2 | 0.50 | 0.73 | 0.52 | 0.65 | 0.33 | x | x | |
| Fr.3 | −0.11 | 0.33 | 0.47 | 0.15 | 0.38 | 0.18 | 0.80 | |
Table 5.
The registration results related to the Patient 5 at different gantry angles: the R-value related to the CC displacement using only kV images as well as kV-MV image pairs, and R-value related to the AP displacement when using kV-MV image pairs. The image sequences for some gantry angles were not available which is specified with x in the table. Cases with R-values higher than 0.5 included P-values < 0.001 which are shown in Bold format. P., G. A and Fr. stand for patient number, gantry angle and fraction, respectively.
|
P.5, G.A. R-value | ||||||||
|---|---|---|---|---|---|---|---|---|
| Fr. | 5° | 40° | 205° | 240° | 280° | 310° | 340° | |
| kV (CC) | Fr.1 | 0.47 | 0.64 | 0.68 | 0.89 | 0.92 | −0.79 | 0.31 |
| Fr.2 | 0.32 | 0.77 | 0.64 | 0.81 | 0.94 | 0.59 | 0.53 | |
| Fr.3 | 0.68 | 0.61 | 0.58 | 0.77 | 0.91 | 0.72 | 0.45 | |
| kV-MV (CC) | Fr.1 | 0.65 | 0.67 | 0.83 | 0.91 | 0.78 | 0.61 | 0.84 |
| Fr.2 | 0.80 | −0.56 | 0.73 | 0.86 | 0.83 | 0.54 | 0.86 | |
| Fr.3 | 0.69 | 0.53 | 0.70 | 0.86 | 0.93 | 0.52 | 0.89 | |
| kV-MV (AP) | Fr.1 | 0.47 | −0.60 | 0.72 | 0.59 | 0.22 | 0.23 | 0.61 |
| Fr.2 | 0.47 | −0.25 | 0.61 | 0.76 | 0.59 | 0.15 | 0.62 | |
| Fr.3 | 0.64 | −0.50 | 0.53 | 0.62 | 0.06 | 0.20 | 0.81 | |
In patient 1, we observed significant R-values for CC displacement at all gantry angles except 58° and 160° when using kV sets for almost all fractions. In addition, at all gantry angles except 160°, significant R-values were achieved for CC displacement when using kV-MV image pairs for all fractions. The registration results were improved when using kV-MV images compared to only kV images where displacement at the gantry angles 25°, 58° showed significant R-values in more fractions. Using the kV-MV image pairs, displacement in AP direction was also successfully extracted for all gantry angles (for almost all fractions) except for gantry angles 130°, 160°, and 190°. For this patient, tumor motion could not be tracked for gantry angle 160° using any of the image sets in any fractions. In patient 2, a significant R value for CC displacement was obtained at all gantry angles except 80° when kV sets were used for almost all fractions. In addition, at all gantry angles except for 120° significant R-value was achieved for CC displacement when using kV-MV image pairs for almost all fractions. The registration results were improved by using kV-MV images compared to only kV images as displacement at the gantry angles 12° and 80° showed significant R-values in more fractions. Using the kV-MV image pairs, displacement in the AP direction could also successfully extracted for all gantry angles except 12° for almost all fractions.
For patient 3, a significant R-value for CC and AP displacement was obtained for all gantry angles when both kV and kV-MV image pairs were used for almost all cases. Good agreement between the reference motion and the extracted displacement is also seen for all cases in Fig. 5. For patient 4 at all gantry angles except for 80° significant R-value was achieved for CC displacement when using kV sets for almost all fractions. In addition, at all gantry angles, R-value was achieved for CC displacement when using kV-MV image pairs for all fractions. The registration results were improved by using kV-MV images compared to when using kV images alone in which displacement at the gantry angles 80°, 170° showed significant R-values in more fractions. Using the kV-MV image pairs, displacement in AP direction was also successfully extracted for gantry angles 80°, 110°, 140°, 230° for almost all fractions.
For patient 5 at gantry angles 40°, 205°, 240°, 280°, 310° significant R-value was achieved for CC displacement when using kV sets for all three fractions. The registration results were improved by using the kV-MV image pairs where the gantry angles 5° and 340° showed also significant R-values. Using the kV-MV image pairs, displacement in AP direction was also successfully extracted for gantry angles 40°, 205°, 240°, and 340°. For patient 4 and patient 5 using kV-MV image pairs, CC translation could be extracted for all gantry angles and all fractions.
4. Discussion
Intrafractional tumor motion, especially in lung tumors, must be considered to reduce the required PTV margin as well as dose delivery to surrounding healthy tissue. To our knowledge, this study is the first to investigate the feasibility of 2D/3D image registration at arbitrary angles for tracking tumor motion using a patient dataset in thoracic radiotherapy.
Two pairs of orthogonal image data sets are used in this study to perform registration in order to obtain CC and AP displacement at various gantry angles. Our results showed that 2D/3D registration can extract tumor motion in 5 DOF using kV images and in 6 DOF using kV-MV image pairs for the majority of gantry angles in all patients. In addition, our results showed that the use of kV-MV image pairs could lead to more successful extraction of CC and AP displacement in most of gantry angles of almost all patients. Tumor motion could not be tracked for the 160° gantry angle with any image set and fraction for patient 1. Our further investigations revealed that at this gantry angle, the tumor is not visible in both kV and kV-MV image pairs, which may be the reason for the failed registration.
We also evaluated the impact of additional MV images in different cases based on tumor visibility in kV and MV image sequences. It was observed that successful registration required visibility of the tumor in at least one set of the image sequences. In cases where the tumor was not visible in any of the image sequences, registration usually failed. However, there were some cases (e.g., for patient 1 gantry angle 25° and 190°) in which the tumor was mostly covered by surrounding structures and not distinctly visible in both image sets, the motion displacement was still successfully extracted. We concluded that the 2D/3D image registration for tumor tracking is feasible for gantry angles mainly in cases where no major other structures overlay the tumor. We should note that the diaphragm motion was only used as our reference motion signal for the evaluation of performance of the registration results in the research stage and would not be needed in future clinical implementation of the algorithm.
For some cases (e.g. patient 5 at gantry angles 205° and 340°) we observed good tumor visibility in MV images and this can be the reason for the improved registration result at 6 DOF. We also observed in some cases that adding the MV images did not improve the registration result (e.g patient 1 gantry angle 98° and 130°). Our further investigation revealed that the tumor was visible only in kV images, whereas no tumor was visible in MV images. This resulted in a good registration result in only 5 DOF, and adding MV images did not improve the results for CC displacement. Data from patient 2 at gantry angle 120° was an exception with a low correlation coefficient despite good tumor visibility in both kV and MV image sequences. We assume the possible reason could be the existence of a high-intensity structure seen in the background which leads the registration to wrongly track the high-intensity structure instead of the tumor.
We observed that for patients 4 and 5 using kV-MV image pairs the CC displacement using kV-MV image pairs has been successfully extracted in all fractions and gantry angles. The tumors in these patients were in the vicinity of the diaphragm (Fig. 1) and therefore a more dominant craniocaudal movement exist which can be the reason for better performance of the registration. By using the kV-MV image pairs and performing registration in 6 DOF, the AP displacement has been successfully extracted for many gantry angles and fractions whereas when using only kV image sets no AP displacement could be extracted (Fig. 6). Patient 4. gantry angle 200°, and patient 5, gantry angle 310° are the only registrations that failed to extract AP motion in every fraction. For Patient 2 we observed a good AP displacement extraction in almost all angles and fractions. Our visual inspection revealed that AP direction is the dominant axis of motion with displacements up to 1.5 cm in this patient, which may be the reason of more successful extraction of AP motion compared to other patients.
In some patients, we observed different registration performance for different fractions when using the same patient data and the same gantry angles. We assume that the reason for this can be due to many configuration parameters that can affect the performance of a 2D/3D registration, such as the intensity configuration, the number and quality of X-ray images per fraction. In addition, proper correlation of kV-MV image pairs, and registration parameters such as choice of the initial guess for registration are among other influencing factors. The addition of a segmentation step to remove the unwanted structures from the ROI could improve the registration result for some ROIs with high-intensity structures in the background (e.g., patient 2 at a gantry angle of 120°) is considered as a future perspective of this study.
5. Conclusion
It was shown that CC and AP tumor displacement could be successfully extracted in almost all patients for most of the gantry angles using intensity-based 2D/3D registration, confirming the possibility of tumor tracking at arbitrary gantry angles. Furthermore, we found that the additional MV image set improved tumor tracking in the CC direction at some gantry angles in all patients. In addition, extraction of AP displacement at arbitrary gantry angles was also possible when using kV-MV images.
Declaration of competing interest
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.
Acknowledgment
This work has been supported by ACMIT – Austrian Center for Medical Innovation and Technology, which is funded within the scope of the COMET program and funded by Austrian BMVIT and BMWFW and the governments of Lower Austria and Tyrol.
Footnotes
Supplementary data to this article can be found online at https://doi.org/10.1016/j.zemedi.2024.03.004.
Appendix A. Supplementary data
The following are the Supplementary data to this article:
References
- 1.Cho B., Poulsen P.R., Sawant A., Ruan D., Keall P.J. Real-time target position estimation using stereoscopic kilovoltage/megavoltage imaging and external respiratory monitoring for dynamic multileaf collimator tracking. Int J Radiat Oncol. 2011;79(1):269–278. doi: 10.1016/j.ijrobp.2010.02.052. [DOI] [Google Scholar]
- 2.Schweikard A., Shiomi H., Adler J. Respiration tracking in radiosurgery. Med Phys. 2004;31(10):2738–2741. doi: 10.1118/1.1774132. [DOI] [PubMed] [Google Scholar]
- 3.Cho B., Suh Y.J., Dieterich S., Keall P. A monoscopic method for real-time tumour tracking using combined occasional X-ray imaging and continuous respiratory monitoring. Phys Med Biol. 2008;53:2837–2855. doi: 10.1088/0031-9155/53/11/006. [DOI] [PubMed] [Google Scholar]
- 4.Hughes S., et al. Assessment of two novel ventilatory surrogates for use in the delivery of gated/tracked radiotherapy for non-small cell lung cancer. Radiother Oncol J Eur Soc Ther Radiol Oncol. 2009;91(3):336–341. doi: 10.1016/j.radonc.2009.03.016. [DOI] [Google Scholar]
- 5.Ren Q., Nishioka S., Shirato H., Berbeco R.I. Adaptive prediction of respiratory motion for motion compensation radiotherapy. Phys Med Biol. 2007;52(22):6651–6661. doi: 10.1088/0031-9155/52/22/007. [DOI] [PubMed] [Google Scholar]
- 6.Künzler T., Grezdo J., Bogner J., Birkfellner W., Georg D. Registration of DRRs and portal images for verification of stereotactic body radiotherapy: a feasibility study in lung cancer treatment. Phys Med Biol. 2007;52(8):2157–2170. doi: 10.1088/0031-9155/52/8/008. [DOI] [PubMed] [Google Scholar]
- 7.Schweikard A., Shiomi H., Adler J. Respiration tracking in radiosurgery without fiducials. Int J Med Robot. 2005;01(02):19. doi: 10.1581/mrcas.2005.010202. [DOI] [Google Scholar]
- 8.Wu J., Kim M., Peters J., Chung H., Samant S.S. Evaluation of similarity measures for use in the intensity-based rigid 2D–3D registration for patient positioning in radiotherapy. Med Phys. 2009;36(12):5391–5403. doi: 10.1118/1.3250843. [DOI] [PubMed] [Google Scholar]
- 9.Birkfellner W., et al. Multi-modality imaging: a software fusion and image-guided therapy perspective. Front Phys. 2018 [Google Scholar]
- 10.Gendrin C., et al. Monitoring tumor motion by real time 2D/3D registration during radiotherapy. Radiother Oncol. 2012;102–142(2):274–280. doi: 10.1016/j.radonc.2011.07.031. [DOI] [Google Scholar]
- 11.Furtado H., Steiner E., Stock M., Georg D., Birkfellner W. Real-time 2D/3D registration using kV-MV image pairs for tumor motion tracking in image guided radiotherapy. Acta Oncol Stockh Swed. 2013;52(7):1464–1471. doi: 10.3109/0284186X.2013.814152. [DOI] [Google Scholar]
- 12.Munbodh R., et al. 2D–3D registration for prostate radiation therapy based on a statistical model of transmission images. Med Phys. 2009;36(10):4555–4568. doi: 10.1118/1.3213531. [DOI] [PubMed] [Google Scholar]
- 13.Xu H., Brown S., Chetty I.J., Wen N. A systematic analysis of errors in target localization and treatment delivery for stereotactic radiosurgery using 2D/3D image registration. Technol Cancer Res Treat. 2017;16(3):321–331. doi: 10.1177/1533034616664425. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 14.Kuo H.C., et al. A phantom study to evaluate three different registration platform of 3D/3D, 2D/3D, and 3D surface match with 6D alignment for precise image-guided radiotherapy. J Appl Clin Med Phys. 2020;21(12):188–196. doi: 10.1002/acm2.13086. [DOI] [Google Scholar]
- 15.Furtado H., et al. FIRE: an open-software suite for real-time 2D/3D image registration for image guided radiotherapy research. San Diego, California, United States. 2016 doi: 10.1117/12.2216082. [DOI] [Google Scholar]
- 16.Powell M.J.D. Springer US; Boston, MA: 2006. The NEWUOA software for unconstrained optimization without derivatives; pp. 255–297. [DOI] [Google Scholar]
- 17.Blessing M. OC-0060: Workflow automation for ultrafast kilovoltage-megavoltage conebeam CT for image guided radiotherapy. Radio Oncol. 2013;106:S22–S23. doi: 10.1016/S0167-8140(15)32366-5. [DOI] [Google Scholar]
- 18.Di Lena P., Margara L. Optimal global alignment of signals by maximization of Pearson correlation. Info Process Lett. 2010;110(16):679–686. doi: 10.1016/j.ipl.2010.05.024. [DOI] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.