Breast positioning control resulted in substantial improvements of geometric implant stability. With this workflow, the mean amount of button–button distance variations was 0.6 ± 1.0 mm, compared to 1.6 ± 2.7 mm obtained for the previous cohort (Fig. 3a). Only 0.4% and 0.4% of the 485 catheters implanted in the 30 patients showed a length shortening and extension > 5 mm (referring to one button size), respectively, whereas this was the case for 2.7% and 3.5% of the 1758 catheters implanted in the 100 patients treated beforehand. Distance deviations with an amount > 1 mm and > 2 mm were measured for 13.6% and 4.5% (with positioning control) as well as for 43.6% and 18.7% (without positioning control) of the implanted catheters. The number of patients affected by large deviations could be significantly reduced as well (Fig. 3b). When treated with the implemented workflow, only 36.7% and 10% of patients revealed at least one distance variation with an amount > 2 mm and > 5 mm, respectively, compared to 80% and 37% of the previous patients. No patient treated with the workflow showed variations > 12 mm, whereas this was observed for 13% of the previous patients. Thus, button-button distance variations were, applying the chi-square distribution tests mentioned in the “Statistical Analysis” section, significantly reduced in terms of both the number of affected catheters (p < 0.001) and patients (p < 0.001).

Shown in Kaplan–Meier style are the relative fractions of implanted catheters (a) and patients (b) affected by at least one button–button distance variation with an amount larger than the corresponding abscissa value for the patient cohorts treated with and without the breast positioning control workflow. The fractions of dwell positions (c) and patients (d) affected by respective Euclidean dwell position deviations are illustrated as well

Regarding the dwell position deviations, breast positioning control led to stability enhancement as well. The mean amount of corresponding deviations was 1.1 ± 1.0 mm when applying the new workflow and 1.4 ± 1.4 mm for the previous patients (Fig. 3c). As example, 16.4% and 1.7% of the dwell positions defined for the patients receiving breast positioning control showed deviations > 2 mm and > 4 mm, respectively, compared to 22.7% and 4.9% obtained for the patients treated before. In particular, Euclidean deviations were significantly reduced regarding both the number of affected dwell positions (p < 0.001) and of affected patients (p < 0.001). Although corresponding deviations with an amount > 5 mm were still observed in 46.7% of the patients treated with breast positioning control (compared to 56% of the patients treated without), only 6.6% of the patients showed a deviation > 7 mm (Fig. 3d). For the previous cohort, such variations were found for 30% of all patients and deviations up to 14.5 mm occurred. To visualize examples of the observed implant alterations, Fig. 4 shows the geometric and dosimetric variations for a patient with very small and very large implant variations, respectively. Furthermore, corresponding changes for the patients showing the maximum implant alterations of both cohorts are provided as Supplementary Material.

a and b show the planning-CT and control-CT situation, respectively, of a patient with very small geometric changes (dwell position deviations < 1 mm) treated with breast positioning control; c and d show the planning-CT and control-CT situation, respectively, of a previously treated patient with large dwell position deviations (up to 8 mm). In this case, catheters that clearly changed their paths within the breast are exemplarily marked with orange arrows. These changes resulted in an underdosage of some target volume parts. Catheter paths (blue) are in each case shown together with the source dwell positions (red) in the upper sections of each subfigure. The resulting dose distribution is visualized in the lower sections. The target volume is marked orange (dashed) in each case. The 70% (green), 100% (red), 120% (yellow), 150% (cyan), and 200% (dark blue) isodose lines are displayed

Breast positioning control improved implant stability particularly in the lateral to medial parts of the breast. As illustrated in the colormaps (Fig. 5), the lateral breast regions close to the patient arm showed respective standard deviations of the dwell position variations of 0.5–1.2 mm for patients treated with positioning control compared to 1.2–1.5 mm obtained for the previous cohort. However, almost no changes in dwell position deviations were observed in the segments adjacent to chest wall. Here, standard deviations of 1.1–1.5 mm were obtained for both patient cohorts. The coronal breast projections showed in each case only small standard deviations in the most cranial segments of 0.6–1.1 mm and 0.9–1.1 mm for patients with and without positioning control, respectively. The most caudal regions corresponded to the segments most affected by dwell position deviations in both cases.

Provided as colormaps are the standard deviations of the Euclidean dwell position deviations observed in the various breast segments, projected into an illustrative axial and coronal breast view for each cohort

Geometric implant alterations resulted for both patient cohorts in dosimetric instabilities. However, regarding the PTV, dosimetric variations were almost identical for patients treated with and without breast positioning control. The mean amount of coverage index CI (i.e. V100) and D90 changes was 1.8 ± 1.4% and 2.5 ± 1.8% for patients receiving the new workflow, respectively, and 1.9 ± 1.4% and 2.5 ± 1.8% for the previous patients (Fig. 6a, b). Nevertheless, we also observed a trend for eliminating larger coverage index and D90 variations, since only one patient treated with positioning control showed variations > 4.3% and > 5.6%, respectively. For the previous cohort, 7% and 9% of the patients were affected by such coverage index (up to 6.1%) and D90 (up to 8.6%) changes. Slight improvements were also observed regarding the COIN (Fig. 6c). For patients treated with and without positioning control, the mean amount of COIN changes was 1.3 ± 2.1% and 2.4 ± 1.9%, respectively. One patient receiving the new workflow showed variations > 5.3%, whereas this was observed for 11% of the previous patients.

Shown in Kaplan–Meier style are the relative fractions of all patients that were affected by coverage index (a), D90 (b), COIN (c), and 150% isodose volume (d) variations with an amount larger than the corresponding abscissa value. Results are provided for both patient cohorts treated with and without the breast positioning control workflow

The 100% isodose volumes changed on average by 1.0 ± 1.6 ccm and 0.8 ± 1.3 ccm for patients treated with and without breast positioning control, respectively. However, the workflow substantially improved the stability of high-dose areas, since the number of patients with large corresponding variations was markedly reduced (Fig. 6d). For instance, no patient treated with the new workflow showed respective changes of the 150% isodose volumes > 3.4ccm, whereas this was the case for 10% of the patients examined before. The latter revealed a maximum variation of 9.7 ccm. As consequence, only one patient receiving positioning control showed absolute DNR changes > 2.4% compared to 13% of the patients treated prior (absolute change of 10.2% at maximum).

The greatest benefit of breast positioning control was found regarding the exposure to OARs. For instance, no patient treated with the workflow showed absolute D0.2ccm and D1ccm skin dose changes with an amount  > 12.4% and > 9.3% of the prescribed dose, respectively, whereas these were observed for 11% and 8% of the patients treated before (Fig. 7a, b). For the latter, the maximum D0.2ccm and D1ccm changes amounted 63.1% and 17.7% of the prescribed dose, respectively. Note, that we are reporting the skin dose changes as fraction of the prescribed dose and therefore describe absolute and not relative dosimetric deviations at this point. Hence, regarding both dose metrics, a substantial reduction of large variations and significantly (p ≤ 0.001 in both cases) improved skin exposure stability was achieved. The mean amount of D0.2ccm variations was 4.4 ± 3.0% and 7.4 ± 10.3%, and the mean amount of D1ccm variations 3.3 ± 2.8% and 4.6 ± 3.4% for patients treated with and without positioning control, respectively. The ESTRO D0.2ccm and D1ccm skin dose constraints [8] were missed by no patient receiving the new workflow on the C‑CT, but by 12 and 4 patients treated without it.

Shown in Kaplan–Meier style are the relative fractions of all patients that were affected by absolute D0.2 ccm (a) and D1 ccm (b) skin dose changes as well as by absolute D0.1 ccm (c) and D1 ccm (d) rib dose changes with an amount larger than the corresponding abscissa value. Results are provided for both patient cohorts treated with and without the breast positioning control workflow

Regarding the rib exposure, breast positioning control significantly improved D0.1ccm (p < 0.001) and D1ccm (p < 0.001) stability as well (Fig. 7c, d). Although only small changes of the mean variation amount (from 3.7 ± 3.4% to 3.4 ± 2.3% of the prescribed dose for D0.1ccm and from 3.1 ± 2.6% to 2.6 ± 1.8% of the prescribed dose for D1ccm) were observed after introducing the new workflow, larger dose instabilities could be avoided. For instance, D1ccm variations of > 4% of the prescribed dose were found for 21.4% of the patients receiving breast positioning control, compared to 32% of the previous patients. Furthermore, only 3.3% of the patients (referring to one patient) showed D0.1ccm variations > 6.7% of the prescribed dose, whereas 16% of the patients treated before did.

In summary, with breast positioning control an improved implant stability was achieved. In particular, large variations regarding high-dose areas and OAR exposure could be substantially reduced or even avoided. Dosimetric stability was also of fundamental importance regarding the decision-tree [6] applied to identify patients requiring treatment adaption. In this respect and as shown in the Supplementary Material, 14% and 1% of the patients treated without breast positioning control required re-consideration due to skin dose variations and target coverage loss, respectively. In contrast, for patients receiving breast positioning control, no individual was identified for re-planning. The effort for the assessment of the need for re-planning was reduced as well, since 67% of the patients (compared to only 55% of the previous cohort) did not require a detailed evaluation of dosimetric implant stability because they showed a high geometric implant stability (see Supplementary Material) in first line. The new workflow therefore effectively helped to avoid additional effort during the brachytherapy course, to enable smooth workflows, and to ensure high treatment quality.