Day to day treatment variations of accelerated partial breast brachytherapy using a multi-lumen balloon.

PURPOSE: To evaluate the variations of multi-lumen balloon (MLB)-based brachytherapy from simulation day to treatment day and their dosimetric impacts during accelerated partial breast irradiation (APBI).
MATERIAL AND METHODS: A total of 42 CT images scanned from seven patients were evaluated with regards to daily variation due to of: 1) internal uncertainty: size and shape of balloon, seroma volume; 2) geometrical uncertainty-random: length of each catheter was measured for each fraction (total 70); 3) geometrical uncertainty-systematic: virtual systematic errors were tested by offsetting dwell positions. The original plans (as group A) had a mean value of 96.8% on V95 of the PTV_Eval. Plans were rerun (as group B) such that the mean value of the V95 was relaxed to 90.4%. By applying the reference plan to each daily CT image, variations of target coverage under different sources of error were evaluated.
RESULTS: Shape and size of the balloon had means of < 1 mm decreased in diameter and < 0.4 cm(3) decreased in volume; the mean seroma volume increased by 0.2 cm(3). This internal variation has a mean of < 1% difference for both V90 and V95. The geometrical uncertainty made a mean deviation of 2.7 mm per root of sum of square. It caused the degradations of V90 and V95 by mean values of 1.0% and 1.2%, respectively. A systematic error of 3 mm and 4 mm would degrade both of V90 and V95 by 4% and 6%, respectively. The degradations on target coverage of the plans in group A were statistically the same as those in group B.
CONCLUSIONS: Overall, APBI treatments with MLB based brachytherapy are precise from day to day. However, minor variation due to daily treatment uncertainties can still degrade tumor bed coverage to an unacceptable coverage when V95 of the original plan is close to 90%.

Purpose

Studies of local recurrences in patients without adjuvant whole breast irradiation (WBI) demonstrated that the majority of in-breast tumor recurrences are within the tumor bed [1-4]. These results support that APBI (accelerated partial breast irradiation) could be a possible option to deliver adjuvant radiation therapy after lumpectomy in patients undergoing breast-conserving therapy (BCT). Results from phase I/II/III studies have shown acceptable local control and cosmesis with APBI [5-7]. Another socio-economic reason that makes APBI a popular choice is that the accelerated treatment, which is typically completed within a week, offers a convenient choice to patients with restricted access to radiation therapy facilities.

Tangential beam irradiation of the whole breast is the traditional technique used to spare lung dose in WBI. Conversely, APBI needs to spare normal breast tissue as much as possible such that 3D conformal radiotherapy (3D CRT) instead of the tangential beam arrangement should be applied in external radiation. APBI using brachytherapy (BT) provides an attractive alternative to 3D CRT. APBI uses 3D CRT as well as interstitial multicatheter brachytherapy (IMBT) planned either to cover a 10-15 mm expansion on the excision cavity plus another 10 mm PTV margin (3D CRT) or to cover 10-20 mm of the surrounding breast tissue of the lumpectomy cavity. APBI using IMBT has the longest treatment history and has provided the most treatment outcome, which is still the standard in evaluating newer APBI technique. MammoSite® Radiation Therapy System (MS) (Hologic Inc., Bedford, MA, USA) offers easier intracavitary implementation to the resection cavity by use of a single lumen balloon, and has been demonstrated as a mature APBI technique. The later developed multilumen balloon (MLB) combines the simplified balloon placement with more complex and flexible dosimetry of IMBT technique [8]. The target volume for the balloon based APBI is to 10 mm (less 5 mm from skin) surrounding the inflated balloon. Due to a tighter margin, the variations from daily treatment should be closely monitored.

Treatment uncertainties exist at each stage of the radiation therapy process [9]. In general, patient-specific clinical uncertainties are related to interfraction and interfraction movement of both target and adjacent organs at risk (OARs). These uncertainties are easier to analyze with geometrical uncertainties and internal uncertainties in BT [10]. Geometrical uncertainties are related to source positioning and to reconstruction of the applicator [11]. Internal uncertainties are caused by potential displacement of the applicator, relative to the target and/or movement of OARs during BT irradiation or in between imaging and dose delivery [12, 13]. AAPM has proposed a task group to develop a consensus on a minimum set of initial QA procedures for clinical implementation of 3D CT-based intracavitary breast brachytherapy. The proposed QA procedures (revised February 13, 2012) includes QA for applicator (physical integrity, source-position accuracy on each lumen, and uncertainty of a catheter length) and QA for CT-based treatment planning (physical dimensions of applicator in CT datasets, applicator reconstruction accuracy, verify DVH, and optimal concentration of image contrast medium in the imaging of applicator). This study aimed to investigate the integrity and variation of daily treatment with MLB. By utilizing CT image-guided BT technique in 4D, daily variations of the MLB implementation were analyzed. The dosimetric impact from these variations was also evaluated.

Material and methods

A total of 42 CT images (one simulation CT plus five daily CTs for each patient) scanned from 7 patients with implanted Contura® Multilumen Ballon (SenoRx, C.R. Bard, Inc., Murray Hill, NJ, USA) were acquired and evaluated in this study. Each patient had at least five 5 mm titanium clips placed at the medial, lateral, superior, inferior, and deep cavity edge (positioned in the center of the cavity) at the time of breast-conserving surgery. Balloon was implanted using ultrasound with close end technique. Depending on the cavity size, different amount of water and 1.5 cm3∼2 cm3 of contrast medium was infused into balloon. Each CT acquisition was done using the following procedures: 1) re-position patient with mold by using lasers against medial and lateral markers, which were marked during simulation; 2) check the orientation of the balloon by aligning lumen #1 with the skin marker drawn during simulation; 3) check the conformance of the balloon to tumor bed by evaluating the air-fluid pocket in CT images; 4) use the vacuum port to aspirate the air-fluid if it is significant and then rescan; 5) measure and record the treatment length of each lumen of the balloon; 6) reconstruct the 1 mm thickness CT images to 22 × 22 cm field of view (FOV) before exporting CT images to the Oncentra® Planning System (Nucletron, an Elekta company, Stockholm, Sweden) for planning.

After the balloon was contoured in each CT image, a PTV_Eval was created using a 10 mm expansion from the balloon, then excluding the balloon volume, the chest wall, and 5 mm from the skin surface. The volume of the 10 mm expansion from balloon was designated as target volume, where target points were generated on its surface for further dose point optimization. An inner skin surface, which was the 2 mm inner distance from external surface [14, 15], and the rib closest to the balloon were delineated to measure the maximum skin dose, (Dskin), and maximum dose to rib, (Drib), respectively. The maximum dose reported in this study was the 0.1 cm3 of the corresponding volume. Planning goals of the original (reference) plan were 95% prescription dose (PD) to more than 95% of the VPTV_Eval (PTV_Eval volume) – V95 > 95%, Dskin < 125% PD, Drib < 125% PD, the VPTV_Eval received 150% PD (V150) < 50 cm3, and the VPTV_Eval received 200% PD (V200) < 10 cm3. A plan with 90% PD to less than 90% of the VPTV_Eval – V90 < 90% was considered unacceptable. The plan was created using dose point optimization in a volume model and followed by manually adjusting the isodose lines (graphic optimization) to achieve the planning goals. In dose point optimization, each point dose is calculated based on TG-43 with anisotropic correction; the dwell time to each source position is the analytical solution solved with singular value decomposition (SVD) such that each point receives the reference dose (3.4 Gy/fraction). By interactively updating dose distribution in 3D and DVH, the graphic optimization allows altering the dwell time weights by locally dragging the isodose lines to adjust the hotspot in the organ at risk and cold spot in target, until the best DHV coverage of PTV_Eval is reached without jeopardizing the other planning goals. This two-step optimization (dose point optimization followed by graphic optimization) has been tested and compared with another IPSA (inverse planning with simulated annealing) optimizations [16, 17]. Since the results from IPSA were not optimum and needed further graphical adjustment to the iso-dose line, we decided to stay with dose optimization in 3D model for APBI brachytherapy with MLD.

The approved plan generated from the simulation CT (reference CT) was saved as reference plan. The treatment length of the MLB measured on the simulation date was recorded as reference length. Before the AM fraction of each treatment day, the entire CT acquisition above (1-5) was repeated. Before the PM fraction of each treatment, procedures (1,2,5) were repeated without CT scan. There were total 42 CT image sets acquired (includes 7 sets form simulation) and 77 treatment lengths recorded (includes 7 reference lengths).

Evaluation of variations due to internal uncertainties (INTR)

In each CT image set, balloon size (Vballoon), diameters of the balloon in AP, and Lat directions (dAP, dLat), the closest skin-to-balloon distance (dskin), the closest balloon-to-rib distance (drib), and the volume of air-fluid pocket (Vseroma), were measured and recorded. Daily internal variations were measured by comparing the differences between the Vballoon, dAP, dLat, dskin, drib, and Vseroma in daily CT image with those in simulation CT image in terms of ΔVballoon, ΔdAP, ΔdLat, Δdskin, Δdrib, and ΔVseroma, respectively. Conformance of balloon to the resection cavity was ensured, as long as Vseroma < 2% or 2 cm3 of the PTV_Eval volume (VPTV_Eval). Its impact to V95 and V90 was accounted by subtracting its volume (in% of VPTV_Eval, if more than 0.5 cm3) from the V95 and V90.

Evaluation of variations due to geometrical uncertainties (GEOR & GEOS)

Due to the variations of the patient positions and the measured conditions between treatments, the reading of separate measured treatment lengths could be different. Each catheter length was measured and recorded for each fraction (total of 70 fractions). Since there are five lumens in a Contura (SenoRx, C.R. Bard, Inc., USA) balloon, the deviation of the catheter length in each fraction is reported by the root of sum of square:

where Δl is the deviation of the length of catheter “i” from its first measured length. By repopulating the time weight of each dwell position of the reference plan onto the corresponding dwell position reconstructed from each daily CT image, dose variations to PTV_Eval, tumor bed (surgical clips), skin, and rib were evaluated with and without daily catheter length corrections. The deviation of the catheter length from the first measurement during simulation was corrected by rounding to the nearest millimeter then offsetting the dwell positions of the corresponding catheter in CT image acquired from treatment day. The variation of catheter length measured from day to day was considered as random variation. The impact of the systematic errors related to source poisoning, treatment length measuring tool, and reconstruction of the catheters in planning system were accounted and tested by systematically offsetting the entire dwell positions of each catheter with ± 1 mm, ± 2 mm, ± 3 mm, and ± 4 mm.

Evaluated parameters

Dosimetric impact was measured in terms of ΔV90, ΔV95, ΔV150, and ΔV200. Dose change in each surgical clip was measured as the minimum dose received by 50% volume of the contoured surgical clip. Dose deviation was compared with and without daily treatment length corrections. ΔV95 and ΔV90 were compared with different type of variations, include daily internal variation from CT image only (INTR), plus daily treatment length variation (GEOR), or plus systematic error with different entire treatment shift (GEOS). Both of GEOR and GEOS were measured from treatment CT, the results inherently included the effect of INTR.

Relaxed plan and statistical analysis

To simulate the dosimetric impact of the plans that did not meet planning goals due to too small skin distance (e.g., dskin < 5 mm), each plan was relaxed to V95 = 90% by graphically dragging the iso-dose line of the original plan inner away from breast tissue. The results from the original plans were grouped as A and the results from the relaxed plans were grouped as B. Dosimetric impact due to daily variations on MLB was compared between group A and group B by two-sided paired sample t-test on the parameters of ΔV95, ΔV90, ΔV150, and ΔV200. It is considered a significant difference when p value < 0.05.

Results

Internal variations (INTR)

Table 1 lists the initial anatomic parameters from the simulation CT image set and its corresponding plan evaluated parameters. Comparing 7 simulation CT image sets with 35 daily treatment CT image sets, dAP and dLat of the balloon, which had initial mean values of 40.9 mm AP and 45.4 mm Lat, had a mean decrease of 0.9 mm and 0.2 mm, respectively; Vballoon, which had an initial mean value of 41.8 cm3, had a mean decrease of 0.4 cm3; Vseroma, which had an initial mean value of 0.2 cm3, had a mean increase of 0.2 cm3; dskin and drib, which had an initial mean values of 18.5 mm and 15.2 mm, had a mean decrease of 1.3 mm and a mean increase of 3.1 mm, respectively. A larger variation of daily reading from the initial reading within patient's treatments and between patient's treatments was seen in dskin. The variation of dskin from day to day for each patient and its corresponding dosimetric impact due to the variation were illustrated in Figure 1. The changes of the relationship between balloon and anatomy making VPTV_Eval, which had an initial mean value 90.9 cm3, had a mean decrease of 1.7 cm3. The overall dosimetric impact of these internal variations had a minor impact on ΔV90, ΔV95, ΔV150, and ΔV200 with mean values of –0.3%, 0.1%, 0.8 cm3 and 0.7 cm3, respectively.

Table 1

Summary of the initial anatomic parameters from the reference CT image set and its corresponding plan evaluated parameters (group A) for each patient in this study. Values within the parenthesis were plan parameters for group B

Patient	dAP1	dLat1	Vballoon2	VPTV_Eval2	Vseroma2	dskin1	drib1	V953	Dskin3	Drib3
1	51	53	69.4	120.2	< 0.1	14	24	100 (90.9)	80.7 (74.9)	55.8 (47.6)
2	35	40	27.3	75.6	< 0.1	14.6	16.4	99.5 (90.1)	75.5 (68)	61.4 (57.1)
3	38	43	31.1	79.7	0.1	33	15	97.9 (90.5)	29.0 (27.4)	63.1 (39.8)
4	42	55	48.3	101.5	< 0.1	22	4	95.0 (90.6)	55.4 (52.2)	116.3 (108.1)
5	42	41	37.3	87.2	< 0.1	13	38	97.7 (90.3)	97.6 (88.5)	29.7 (27.8)
6	45	50	55.1	107.1	< 0.1	12	5	95.1 (90.0)	101.4 (95.0)	121.9 (117.8)
7	33	36	24	65	1.1	21	4	92.4 (90.1)	61.6 (61.1)	122.4 (117.3)
Mean	40.9	45.4	41.8	90.9	0.2	18.5	15.2	96.8	71.6	81.5

Unit: 1 – mm; 2 – cm3; 3 –%.

Fig. 1

Scatterplot of the variation of dskin due to INTR and the corresponding Dskin, where “% PD” was the percentage of the prescription dose and the larger symbols were the intial values taken from reference CT

Geometrical variations (GEOR & GEOS)

In addition to INTR, Table 2 demonstrates the dosimetric impact of GEOR and GEOS on PTV_Eval. The variation of the treatment lengths of the five lumens in terms of RSQL had a mean value of 2.7 mm. On top of INTR, this deviation caused further degradation of the PTV_Eval coverage and an increase of the hot spot within PTV_Eval for both of group A and group B. There were mean values of –1%, –1.2%, 0.8 cm3, and 1.0 cm3 for ΔV90, ΔV95, ΔV150, and ΔV200 of group A, respectively. The impacts of INTR & GEOR on the mean dose (which is close to 50% DVH value) of each surgical clip are displayed in Figure 2. Although the statistical normality test did not suggest the frequency distributions follow a normal distribution, which may be due to too small of a sample set, both distributions showed the impacts were random.

Table 2

Summary of internal variations due to the change of balloon and anatomy, geometrical variations due to the change of catheter length, size of systematic error, and their dosimetric impact on PTV_Eval

	ΔdAP1	ΔdLat1	ΔVballoon2	ΔVseroma2	Δdskin1	Δdrib1	ΔV903	ΔV953	ΔV1502	ΔV2002	RSQL3
*INTR	–0.9 ± 2.1	–0.2 ± 1.1	–0.4 ± 1.6	0.2 ± 0.9	–1.4 ± 9.6	3.1 ± 7.0	–0.3 ± 2	0.1 ± 2	0.8 ± 2.7	0.7 ± 2
$INTR							–0.3 ± 2	0.0 ± 2	0.8 ± 2.7	0.7 ± 2
*GEOR							–1.6 ± 2	–1.2 ± 2	0.8 ± 2.7	1.0 ± 2.8	2.7 ± 1.6
$GEOR							–1.0 ± 2	–1.3 ± 2	0.7 ± 2.8	1.0 ± 3
GEOS	S–4mm	S–3mm	S–2mm	S–1mm	S1mm	S2mm	S3mm	S4mm
*ΔV901	–4.0 ± 8	–2.1 ± 6	–0.7 ± 4	–0.0 ± 2	–0.7 ± 2	–1.9 ± 3	–3.6 ± 5	–5.6 ± 7
$ΔV901	–5.6 ± 13	–3.4 ± 11	–1.6 ± 8	–0.5 ± 4	–0.9 ± 4	–2.6 ± 7	–4.9 ± 7	–7.7 ± 8
*ΔV951	–5.2 ± 11	–2.8 ± 9	–1.1 ± 7	0.0 ± 4	–0.8 ± 3	–2.4 ± 5	–4.4 ± 7	6.7 ± 9
$ΔV951	–4.4 ± 13	–2.1 ± 11	–0.5 ± 8	0.5 ± 5	0.0 ± 4	–1.6 ± 7	–4.0 ± 8	–6.8 ± 8

Unit: 1, mm; 2, cm3; 3,%. Value displays as Mean±2SD.

Group A

Group B

Fig. 2

Frequency distribution of the change of 50% dose of sugical clips due to INTR and GEOR, in terms of the percentage of priscription dose

The lower half of Table 2 summarizes the degradation of the PTV_Eval coverage related to GEOS (on top of INTR). In terms of ΔV90 for group A, they were –0.0% and –0.7% for ± 1 mm, –0.7% and –1.9% for ± 2 mm, –2.1% and –3.6% for ± 3 mm, and –4% and –5.6% for ± 4 mm. In terms of ΔV95 for group A, they were –0.0% and –0.8% for ± 1 mm, –1.1% and –2.4% for ± 2 mm, –2.8% and –4.4% for ± 3 mm, and –5.2% and –6.7% for ±4 mm. The values of ΔV90 and ΔV95 in group B had similar mean and variance compared to those in group B.

Comparing original plans in group A, which has a mean V95 of 96.8%, and the relaxed plans in group B, which has a mean V95 of 90.4%, using the testing parameters INTR, GEOR and GEOS (± 1 mm, ± 2 mm, ± 3 mm, and ± 4 mm), the p values (median and range) for ΔV90, ΔV95, ΔV150, and ΔV200 were 0.25 [0.04, 0.64], 0.58 [0.10, 0.93], 0.04 [0.01, 0.28], and 0.29 [0.23, 0.94], respectively. It indicated that only ΔV150 (on GEOS with ± 2 mm, ± 3 mm, and ± 4 mm) showed a significant difference between these two groups. Since the majority of differences were not significant, the statistics display shown in Figure 3 were evaluated including both groups. It showed that the INTR had the least dosimetric impact, followed by GEOS-1mm, GEOR, GEOS-2mm, GEOS-3mm, and GEOS-4mm. In terms of the percentage of plans that became unacceptable plans (V90 < 90%) due to daily treatment variations in this study, in group A, the chance was less than 3%, even if there was a 3 mm systemetic error since its baseline (mean V95 = 96.8%) was high. However, in group B, the chance was more than 14% and 34% if the systematic error was more than 2 mm and 3 mm, respectively.

Fig. 3

Boxplots of the dosimetric impacts (ΔV90, ΔV95, ΔV150, ΔV200) vs. different type of variations, where INT is INTR, GEO is GEOR, S represents GEOS and is followed with different size of systematic errors

Discussion

In an effort to analyze the day to day variation of APBI in patients implanted with MLB, this study used daily CT and pre-treatment catheter length measurements to demonstrate the possible variations due to internal variation and geometrical uncertainties. The major variations of an intracavitary balloon implant include the conformance of the balloon to the resection cavity, shape of the balloon, distance of balloon to OARs, and edema of the breast tissue around the tumor bed due to the implant. It has been reported that a high percentage of implanted patients had balloons explanted due to poor cavity-balloon conformance, or inadequate skin spacing [18, 19]. Contura® (SenoRx, C.R. Bard, Inc., USA) MLB has a vacuum port, which can be used to aspirate the air-fluid pocket if it is significant. Studies have shown that some patients are likely to produce seroma after surgery or implantation [20]. Bhatt [21] observed a trend for larger seroma accumulation if the volume drawn on the first day of treatment was more than 3-6 cm3, comparing to those of 0-2 cm3. Their data suggested that serial aspiration may benefit a subpopulation of patients in terms of tissue conformality and reproducibility. Two out of seven patients in this study needed to aspirate the air-fluid pocket a couple of times in five treatment days. After aspiration, only in one CT image the air-fluid pocket had volume slightly more than 2 cm3, which is less than 3% of the PTV_Eval volume. The variation of the balloon in terms of shape (diameters) and size (volume) were all small. dskin was 1.4 mm smaller and drib was 3 mm larger than the distances measured in the reference CT. Edema due to implant of the balloon is difficult to identify with a CT image. With the trend of the increasing drib over time, it could be an indication of the release of the edema.

Dosimetric impact due to MLB rotation has recently studied by Kim et al. [22]. For a plan with both of skin and rib distance smaller than 7 mm, the maximum degradation on D95 of the PTV_Eavl was about 6.5% when there was a 120° rotation error. They found that in a clinical situation, which had a rotation error smaller than 10°, the degradation on D95 and D90 of the PTV_Eval were less than 1%; the deviation of the skin and rib dose were less than 2.5% of the prescription dose. The geometrical variation due to the orientation of the MLB within the cavity of the lumpectomy had been implicitly evaluated within the CT image. It was first visually inspected by aligning the line on the shaft of Contura (SenoRx, C.R. Bard, Inc.) with the line drawn on patient at the time of initial CT. Ouhib [23] confirmed that visual inspection with marker alignment is an effective way of verifying the orientation of the MLB before treatment. Our experiences showed that with the shaft of the balloon aligned to the external marker, it should still have minor misalignment internally compared to initial CT image, since both of the shaft of Contura and breast tissue are non-rigid. Including this deviated balloon orientation with the other internal uncertainties between balloon and internal anatomy, the INTR in our study showed minimal deviation from the initial plan whether comparing the impact on the tumor bed surrogate, the surgical clips, or directly on the PTV_Eval. The spatial variation of the surgical clips over treatment course has been presented and will be reported separately.

A 2.7 mm RSQL was not small if compared to the 2 mm source position tolerance implemented in most HDR QA programs. The cause of the 3 mm length deviation is not clear. We have observed that the bending and rotation of the middle soft flexible tube caused variations in lumen length. The length could have a deviation of up to 4 mm if the balloon was rotated by 90° and the lumen bent by 60° at the same time. Fortunately, the directions and magnitudes of the deviation varied between the five lumens and usually had an arithmetical mean of less than 1.5 mm. By comparing the overall effect of this GEOR with GEOS in Figure 3, the mean 1% degradation of the PTV coverage due to GEOR was close to GEOS of ∼1.5 mm. The dosimetric impact caused by GEOR was slightly worse than the impact of INTR, however, the range of the error induced was tolerable in this study. In contrast, GEOS had a larger dosimetric impact compared to INTR and GEOR. The impact increased with the systematic error and the probability that the original plan became unacceptable was proportional to the size of the error. If there were a systematic error of 3 mm, the mean degradation of V95 was ∼4%, which would make the majority of the plans in group A became minor deviation from the planning goal and turn more than 30% of the plans in group B into unacceptable plans.

The geometrical uncertainties analyzed in this study were applied to a PTV volume with a standard 10 mm margin from the implanted balloon, which was estimated to be equivalent to treatment thickness of 16 mm and 20 mm in Dickler and Weed's studies, respectively [24, 25]. In a different study [26], Vicini et al. concluded that the effective volume expanded from the 10 mm MammoSite CTV margin was significantly smaller than the 3D-CRT CTV based on a 15 mm margin. Besides, definition of the PTV volume was done irrespective to the status of the surgical margin. Studies showed that the majority of the patterns of failure after APBI were out of field recurrence or elsewhere recurrence [27-30]. However, it was confirmed that close-margin or close/positive-margin have a significant increase in IBTR (in breast tumor recurrence) compared to negative-margin [31]. According to the statistics measured by Husain [27], the number of BT cases increased nearly 10-fold from 2002 to 2007, and more than 60% of the patients who received APBI via BT would fall into ASTRO's cautionary or unsuitable groupings. There is no direct relationship between the patterns of failure and planning coverage, however, it is good practice to plan with objective to meet the goal of V95 > 95%, and to follow a proper QA procedure for a balloon-based APBI patient who is defined as a cautionary or unsuitable APBI BT patient by ASTRO.

This study did not intent to evaluate the variation due to the delineation of the balloon and critical structures, however, efforts have been done to reduce the variation. Each balloon has been infused with about 2 cm3 contrast to increase the visibility of the balloon. Each CT was scanned with 1 mm thickness and reconstructed with 22 cm × 22 cm field of view to increase the resolution of the CT image. Albeit, a small number of patients, this study increased the sample size by relaxing plans with 5% lower than planning goal such that both groups can be tested via day to day uncertainties. As well as small sample size, another limitation of this study is that the impact of day-to-day variations on Dskin and Drib was not intensively studied and documented, since both of dskin and drib in our study were not short enough to receive more than 125% PD after uncertainties were applied. This preliminary study showed that drib tended to be increased from the initial value while dskin did not have similar trend. The fitting curve in Figure 1 indicated that small initial dskin had higher dosimetric impact than large initial dskin. A mean decrease of 2 mm dskin from the reference CT image could have a 20%∼40% dose increase in treatment CT image when the initial dskin is smaller than 12 mm. This would greatly increase the risk of skin complication if the initial Dskin were close to 125% of the prescription dose. This study did not include case with dskin smaller than 10 mm. A further study for cases with short skin distance and rib distance is warranted. Although plans were created using a Contura® (SenoRx, C.R. Bard, Inc.) MLB implant, the conclusions can be applied to any balloon based implant including MS (Hologic Inc.). MS (Hologic Inc.) does not need to consider balloon orientation, however, there is no vacuum port, so the volume of the air-fluid pocket has to be monitored. Clinical outcomes of MS (Hologic Inc.) treatments have shown similar recurrence rates to IMBT, but the late toxicity of MS (Hologic Inc.) treatments were much higher than those of IMBT treatments [31]. MLB have no shortcomings in MS (Hologic Inc.) and have the advantages of both MS (Hologic Inc.) and IMBT, making MLB much more applicable than MS (Hologic Inc.) for APBI. However, the treatment history of MLB is still short and it takes 2-5 years for the incidence of recurrence and late toxicity. APBI with MLB needs to be implemented with great care.

It is the purpose of this study to have more understanding on the variation from MLB such that a proper QA procedure can be followed. Based on the results of this study, we suggest: 1) setup position should be reproduced and MLB orientation should be checked at each fraction; 2) the measured condition applied to MLB should be as close as possible to the MLB status during treatment with minimal bending and twisting of the MLB; 3) the reference treatment lengths measured during simulation should be confirmed at least before the first treatment; 4) ultrasound or CT scan should be checked at least at the first fraction to confirm the integrity of the implanted balloon and no new major air-fluid formation; 5) if serious air-fluid were found, use vacuum port to aspirate the air-fluid; a CT scan needs to be repeated in the following day until no new air-fluid is formed; 6) if air-fluid cannot be eliminated and is more than 5% of VPTV_Eval, re-plan with new CT or explant the balloon if new plan is not acceptable.

Conclusions

For MLB-based APBI patients whose treatment positions are reproducible, the day to day variation on balloon size, skin distance, and rib distance has very minimal impact on V90, V95, V150, and V200. Deviation of measured catheter length from the first day measurement had a mean RSQL of 2.7 mm. Combining the measured deviation of catheter length and the variation from the internal position, V95 was degraded by –1.2 ± 2.3%. If there were a systematic error of 3 mm, the mean degradation of V95 was about 4%. It would turn the majority of plans, which has no deviation from the goal of V95 > 95% into minor deviation of 90% < V95 < 95%, and turn more than 30% of the plans with original V95 close to 90% into an unacceptable plan of V90 < 90%. A proper work flow is essential for MLB-based APBI treatment. A 2 mm tolerance of a HDR QA program is reasonable for regular MLB implant technique for plans with no or minor deviation from the planning goals. Plans with major deviation (e.g. V95 < 90%) should consider to re-implant with different size of balloon or to treat patient with other technique.