Comparing Digital Tomosynthesis to Cone-beam CT for Position Verification in Patients Undergoing Partial Breast Irradiation

Abstract

Purpose

To evaluate digital tomosynthesis (DTS) technology for daily positioning of patients receiving accelerated partial breast irradiation (APBI) and to compare the positioning accuracy of DTS to 3D cone-beam computed tomography (CBCT).

Methods and Materials

Ten patients who underwent APBI were scanned daily with on-board CBCT. A subset of the CBCT projections was used to reconstruct a stack of DTS image slices. To optimize soft-tissue visibility, the DTS images were reconstructed in oblique directions so that the tumor bed, breast tissue, ribs, and lungs were well separated. Coronal and sagittal DTS images were also reconstructed. Translational shifts of DTS images were obtained on different days from the same patients and were compared to the translational shifts of corresponding CBCT images. Seventy-seven CBCT scans and 291 DTS scans were obtained from 9 evaluable patients.

Results

Tumor beds were best visible in the oblique DTS scans. One-dimensional positioning differences between DTS and CBCT images were 0.8 to 1.7 mm for the six patients with clips present and 1.2 to 2.0 mm for the three patients without clips. Due to the limited DTS scan angle, the DTS registration accuracy along the off-plane direction is lower than the accuracy along the in-plane directions.

Conclusion

For patients receiving APBI, DTS localization offers comparable accuracy to CBCT localization for daily patient positioning while reducing mechanical constraints and imaging dose.

INTRODUCTION

In accelerated partial breast irradiation (APBI) treatments, the irradiated volume is relatively small compared to conventional tangential fields, and hence there is particular interest in using 3D imaging techniques to guide patient positioning. On-board cone-beam CT (CBCT) technology provides 3-dimensional (3D) anatomical information for patient positioning and therefore helps to reduce daily set-up error. Recent studies have reported that CBCT imaging can achieve 1 to 2 mm positioning accuracy for APBI setups ^1–3. However, current CBCT techniques may not be optimal for APBI treatments in terms of radiation dose, acquisition time, and geometric clearance. Since the isocenter for breast patients is often placed several centimeters lateral to midline, the couch often needs to be shifted medially (i.e, back to medial position) so that the full gantry rotation can be cleared without collision. Consequently, the contralateral breast and lung receive the same x-ray imaging dose (5~ 8 cGy) as the ipsilateral breast and lung⁴. To protect healthy tissue and minimize secondary cancer occurrences, the dose to the contralateral breast and lung should be minimized if possible ^5–7.

Digital tomosynthesis (DTS) is a new on-board imaging approach that has been proposed to substantially reduce gantry rotation angles for on-board 3D imaging ⁸. Unlike CT scans, which require clearance of 360° gantry rotation, DTS scans acquire projections over a small scan angle (e.g., 45° or less). Potentially for peripheral targets such as the breast, no couch shift may be needed because the gantry only needs to rotate over a small angle. As a result, both imaging dose and acquisition time are reduced to a small fraction of those required for CBCT scans. The contralateral breast and lung dose can be minimized because the scanning volume and angle can be selectively limited to the treatment site. Given these practical interests, DTS is a potential 3D-imaging method for image-guided radiation therapy (IGRT) ^8–10.

DTS imaging technique has been studied extensively for several decades ^11–13 and has been implemented in many diagnostic clinical applications ^14–17. Recent works have demonstrated its potential application to IGRT ^8–10. For example, Wu et. al. reported that the positioning discrepancy between DTS and CBCT imaging was less than 1 mm for head and neck patients ¹⁰. The present study addresses the utility of DTS for localization in patients receiving APBI treatments. The purpose of this study is to quantitatively compare translational shifts of DTS images obtained on different days to the translational shifts of corresponding CBCT images. We considered the CBCT-detected shifts as the “gold standard” and used them to evaluate the accuracy of the DTS-detected shifts. DTS images were constructed from coronal, sagittal, and oblique scans independently. Image quality and registration accuracy were compared across the three scans. Registration accuracy was also compared between patients with and without surgical clips. The limitation of DTS imaging was studied by comparing in-plane registration accuracy to off-plane accuracy.

METHODS AND MATERIALS

Patient data

Ten patients receiving external-beam partial-breast irradiation were enrolled in an Institutional Review Board approved study to evaluate the utility of CBCT and DTS imaging in reducing patient set-up variations. Written informed consent was obtained for all patients prior to enrollment. Typically, a 10 to 15 mm margin was placed around the tumor bed to generate the CTV. Another 5-mm margin was typically placed around the CTV to generate the PTV. The CTV margin was consistent with the ongoing phase III clinical trial for APBI (NSABP B-39) and the PTV margin was smaller than the 10-mm margin used in the trial ^{3, 18}. The treatment plan for each patient consisted of 3 to 4 conformal photon beams, using 6 MV or 15 MV set at the geometric center of the PTV and orientated in a non-axial and non-opposed fashion. Patients received 10 fractions of 3.85 Gy each and were treated to a total dose of 38.5 Gy. The 95% isodose line was typically selected as the dose-volume prescription in most plans, at the discretion of the treating physician. In some patients, part of the prescription dose was delivered with an en face electron field, whose setup was not part of the current study.

Initially, 100 CBCT images were to be obtained for the study. However, 19 CBCT scans could not be completed due to technical considerations. One of the ten patients enrolled in the study only had 4 CBCT scans because a software upgrade that coincided with her treatments prevented imaging. For the other 9 evaluable patients, three had 10 CBCT scans, three had 9 scans, two had 8 scans, and two had 7 scans. To minimize statistical variation, the patient with only 4 CBCT scans was excluded from the study. Surgical clips were present at the excision bed in 6 of the 9 evaluable patients and were used for image registrations. No clips were available for the other 3 patients, and tumor beds were used for registration landmarks.

On-board image acquisition

Patients were initially positioned by aligning the skin marks to the lasers. Orthogonal kV/MV images were taken and then registered on digitally reconstructed radiographs (DRR) based on visible bony anatomy (e.g., ribs, lung interface, vertebral column, and sternum). The couch shifts detected by the 2D kV/MV radiographs were then applied to correct the patient position. After the kV/MV radiograph setup, patients underwent a full 360° CBCT scan using the OBI system (Varian Medical Systems, Inc., Palo Alto, CA). Before the CBCT scan, the couch was shifted laterally by 5 to 15 cm (sometimes vertically as well) to prevent collision during gantry rotation. These shifts were recorded and utilized during later image registrations. In order to increase the field of view, a half-fan mode was used for the CBCT scan. The detector was shifted to one side by 15 cm. A half-fan bowtie filter was mounted outside of the kV x-ray tube. For each scan, 650 to 700 projection images were acquired to reconstruct volumetric CBCT images, using the Feldkamp filtered back-projection algorithm ¹⁹. The reconstructed CBCT images had 2.5 mm slice thickness and 0.98 mm pixel size in axial slices. Details of the CBCT system can be found in ²⁰.

DTS image reconstruction

A subset of each CBCT scan was created to include approximately 80 projections over a 45° scan angle. This subset was used to reconstruct a set of 3D DTS images, using the same Feldkamp filtered back-projection algorithm that was used to reconstruct CBCT images ¹⁹. To reduce computational costs, the projections were down-sampled to 512 × 384 pixels prior to reconstruction, resulting in a 0.776-mm pixel size in the detector plane which corresponds to a 0.52-mm pixel size in the isocenter plane. The reconstructed DTS images consisted of 512 × 320 × 400 voxels with a 0.5 × 0.5 × 0.5 mm³ size for each voxel.

In this study, DTS images were reconstructed by using in-house software that took about 20 minutes. CBCT images were reconstructed by using the vendor provided software that took 1–2 minutes. Future DTS reconstruction may be reduced to less than 1 minute by using a hardware acceleration that has been recently implemented at our institution ²¹.

DTS scans from three orientations

DTS scans were generated along sagittal, coronal, and oblique orientations (Figure 1). Sagittal DTS images were reconstructed using the projections between 270°+22.5° and 270°−22.5°. Due to the 15-cm detector shift applied in the half-fan CBCT scan mode, only one side of the breast is visible in coronal and oblique DTS images. Coronal DTS images were reconstructed using the projections between 180°+22.5° (IEC convention) and 180°−22.5° for right breast or the projections between 0°+22.5° and 0°−22.5° for left breast. Oblique DTS images were reconstructed using the projections between 225°+22.5° and 225°−22.5° for right breast and the projections between 315°+22.5° and 315°−22.5° for left breast.

Figure 1.

Diagrams for coronal, oblique, and sagittal DTS scans for right and left breast treatments

Interfraction 3D image registration

The first fraction CBCT image set was registered to each of the subsequent CBCT image sets in order to determine translational shifts between fractions. The registrations were based on the tumor bead and clips, as seen on the axial, sagittal, and coronal views. These interfraction variations are different from the couch shifts performed at the treatment machine, which are detected by registering CBCT images to the planning CT images. The present analysis was not meant to assess the degree of set-up variations, but rather to determine if the detected interfraction shifts were similar for CBCT and DTS images. For this purpose, the first fraction DTS image set was registered to each of the subsequent DTS image sets. Registration between two DTS images consisted of two steps: off-plane and in-plane registrations. In the first step, the DTS images was carefully aligned across the reconstruction planes so that both images showed the best focus for the same clips or soft-tissue landmarks. In the second step, the two DTS images were aligned in the reconstruction planes so that the clips or soft tissue landmarks matched. Our clinical protocol for APBI treatment with daily CBCT guidance specifies that setup corrections are made in translational directions only. Hence, this study only analyzed translational positioning accuracies and not rotational ones.

Three DTS scans (coronal, sagittal, and oblique) were considered independently and the registration of each DTS scan was blinded to the registration of the other scans. When surgical clips were available, the two clips closest to the treatment isocenter were used as landmarks. When there were no surgical clips, the tumor bed itself was used as the landmark. Both the CBCT and DTS registrations were manually performed by both a physician and a physicist. The registrations were performed using the off-line review 3D-registration software supplied by the vendor (Varian Medical Systems, Inc., Palo Alto, CA). Registration results from each observer were analyzed individually, and interobserver variations were evaluated.

Statistical analysis for registration accuracy

The CBCT-detected shifts were considered as the “gold standard”. Registration differences were calculated for each pair of CBCT and DTS registration results. The root-mean-square (RMS) error was used to evaluate the registration accuracy along lateral, vertical, and longitudinal directions and also for the vector sum. The RMS error was calculated as the square root of the mean square of the registration differences over the evaluable data in the whole dataset.

Registration accuracy was compared between patients with and without surgical clips and also across the three DTS scans. For the first comparison, the p-value was calculated using the unpaired Student’s t-test. For the second comparison, the paired Student’s t-test was used to calculate the p-value because the detected shifts in the three DTS scans reflected the same translational shifts.

In-plane and off-plane accuracy

DTS image stacks have two in-plane directions and one off-plane direction. For coronal DTS images, vertical is the off-plane direction while lateral and longitudinal are the in-plane directions. For sagittal DTS images, lateral is the off-plane direction while vertical and longitudinal are the in-plane directions. For oblique DTS images, the off-plane direction is along the gantry-centering angle at either 135° (left breast treatment) or 45° (right breast treatment). The in-plane directions are the longitudinal direction and the radial direction, which is along the gantry angle at either 45° (left breast treatment) or 135° (right breast treatment). Unlike CT images, which have relatively homogenous image resolution, DTS images have lower resolution along the off-plane direction than the in-plane directions, which may result in registration degradation. The level of this degradation was clinically determined by comparing the in-plane and off-plane registration differences. Statistical significance was calculated using unpaired Student’s t-test.

RESULTS

Image quality comparisons across three DTS scans

In general, the DTS images exhibited high spatial resolution in the in-plane views. When present, surgical clips were clearly visible in all three DTS scans. When surgical clips were not present, the tumor beds were consistently visible only in the oblique scans. As shown in Figures 2 and 3, it was difficult to identify the tumor bed in one or both of the coronal and sagittal DTS images in some patients because bone and breast tissue may have added strong shading to the tumor bed. In the oblique scan, the breast tissue, bones, and lung were well separated, and the soft-tissue contrast of the tumor bed was good enough to register DTS images. Therefore, only oblique scans were used for patients without clips.

Figure 2.

Three views of CBCT and DTS for patient A. While the surgical clips were visible in all three DTS scans, the tumor bed was clearly visible only in coronal and oblique scans.

Figure 3.

Three views of CBCT and DTS for patient B who had no surgical clips implanted. The tumor bed (indicated by the arrows) is not visible in the coronal scan and is barely visible in the sagittal scan, but is clearly visible in the oblique scan.

Positioning differences between DTS and CBCT

Root-mean-squares (RMS) were calculated for the positioning differences between the DTS and CBCT registrations (Table 1). The results are listed for two separate groups: Group A consists of the 6 patients with surgical clips present and Group B consists of the 3 patients without clips. For Group A, the RMS of the positioning difference ranged from 0.85 to 1.67 mm along the lateral, vertical, and longitudinal directions, and 1.80 to 2.45 mm for the vector sum. For Group B, the tumor bed was only visible in the oblique scans. The RMS of the positioning difference ranged from 1.24 to 2.03 mm along the three directions and 2.78 mm for the vector sum.

Table 1.

RMS values of the positioning differences between the DTS and CBCT registrations. Units are in mm.

	Group A: 6 patients/52 fractions based on surgical clips			Group B: 3 patients/25 fractions based on tumor bed
	Coronal DTS	Sagittal DTS	Oblique DTS	Oblique DTS
Lateral	0.90	1.67	1.29	2.03
Vertical	1.31	1.28	1.35	1.24
Longitudinal	0.85	1.25	1.10	1.43
Vector Sum	1.80	2.45	2.17	2.78

Comparisons between patients with and without surgical clips

DTS registrations based on the tumor bed had significantly lower accuracy than registrations based on surgical clips (Table 1). Group B had comparatively larger RMS’s in the lateral and longitudinal directions than Group A, and also for the vector sum (p = 0.05, 0.19, and 0.03, respectively). The RMS along the vertical direction was not statistically different between the two groups (p=0.30).

Comparisons across three DTS scans

Positioning differences were compared across the coronal, sagittal, and oblique DTS scans for the six patients with surgical clips. The coronal scans had a smaller RMS value for the lateral direction than the sagittal and oblique scans (p = 0.004 and 0.05, respectively). The sagittal scans had a smaller RMS value for the vertical direction than coronal and oblique scans (p = 0.09 and 0.008, respectively). These differences can be explained by the fact that DTS scans have lower resolution in the off-plane direction than the in-plane directions. As mentioned earlier, the lateral direction is inside the view plane for coronal scans but orthogonal to the view plane for sagittal scans, and likewise is the vertical direction for sagittal and coronal scans. However, the longitudinal direction is inside the view plane for all three DTS scans. Our results showed that there was no statistical difference of longitudinal registration accuracy across the three DTS scans (p = 0.56 and 0.89 for the coronal-to-sagittal and coronal-to-oblique comparisons).

The vector sum of set-up variations was the square-root-sum of positioning differences along all the three directions. Coronal DTS had a slightly smaller RMS for the vector sum than sagittal and oblique DTS, but the differences were not statistically significant (p = 0.19 and 0.23, respectively).

In-plane and off-plane accuracy

In-plane and off-plane differences between DTS and CBCT registrations were calculated (Table 2). For the six patients with surgical clips (Group A), the off-plane difference (1.31 to 1.60 mm) was significantly lower than the in-plane difference (0.85 to 1.28 mm) for all three DTS scans (p = 0.0001, 0.01, and 0.0001 for coronal, sagittal, and oblique DTS scans, respectively). No statistical difference was found for the three patients without surgical clips (Group B).

Table 2.

Comparisons of the RMS values of positioning differences between in-plane and off-plane directions. Units are in mm. P-values were calculated for the hypothesis that in-plane errors are smaller than off-plane errors.

		Group A: 6 patients/52 fractions based on surgical clips			Group B: 3 patients/25 fractions based on tumor bed
		Coronal DTS	Sagittal DTS	Oblique DTS	Oblique DTS
In-plane	Longitudinal	0.85	1.28	1.10	1.43
	Axial in-plane	0.90	1.25	0.96	1.75
Off-plane		1.31	1.67	1.60	1.62
p-value		0.0001	0.01	0.0001	0.5

Interobserver Variations

We analyzed the CBCT and DTS registrations from each observer individually. The results presented above are based on the registrations performed by the physicist. The registrations from the other observer (the physician) yielded similar results.

There existed some discrepancies between the registration results performed by the physicist and the physician because they may register the tumor bed in slightly different ways. The RMS of the interobserver variation in the DTS registrations was 1.5, 0.9, and 0.9 mm along the lateral, longitudinal, and vertical directions, respectively. Similar interobserver variations were reported for CBCT breast registrations ¹.

DISCUSSION

Comparisons with head-and-neck treatment setups

Wu et. al. reported that DTS imaging can achieve a positioning accuracy of < 1 mm for head-and-neck treatment setups when bony anatomical structures were used as reference landmarks ¹⁰. In comparison, our study shows that 1 to 2 mm accuracy can be achieved for APBI treatment setups when surgical clips are used as reference landmarks. When soft-tissue structures are used as registration landmarks, the localization accuracy becomes lower (2 to 3 mm)..

Comparisons with orthogonal kV imaging setups

Orthogonal kV imaging is also widely used in patient setups. For example, Fatunase et al. ³ evaluated the accuracy of orthogonal kV imaging setups against CBCT setups and found a 3-to-4 mm discrepancy, which is significantly larger than the discrepancy found in this study between DTS and CBCT. This difference is mostly because the orthogonal kV imaging setups were based on bony structures, while DTS and CBCT setups were based on soft-issue structures.

Limitation of DTS imaging

Our study shows that DTS imaging has comparable positioning accuracy to CBCT. However, DTS images do not have sufficient volume information to monitor tumor volume changes. Therefore, CBCT scans may still be necessary throughout a course of treatment (e.g., on the first day of each week’s treatments). For the remaining treatment days, DTS scans may be used to accelerate the on-board imaging process and reduce X-ray imaging dose.

Data limitations

This study was based on only 9 evaluable patients and 77 evaluable CBCT scans. Having more data would help reduce statistical variation. Another limitation of our study is that we did not use Reference-DTS (RDTS) images, as did in References 8 and 10 ^{8, 10}. This was because the clinical CT slice thickness was 5 mm in our study, which was too thick to generate RDTS images.

CONCLUSION

The positioning difference between DTS and CBCT localization was 1 to 2 mm. For patients receiving accelerated partial breast irradiation, DTS localization offers comparable accuracy to CBCT localization for daily patient positioning while reducing mechanical constraints and imaging dose. These two factors are both clinically and biologically important for APBI treatments in which soft-tissue visibility and protection from x-ray imaging dose to the contralateral breast and lung are crucial.