inter-institute variability of kb-models for whole breast irradiation with tangential field


alessia tudda1, roberta castriconi1, elisabetta cagni2, giovanna benecchi3, francesca dusi4, piergiorgio esposito1, marika guernieri5, anna ianiro6, valeria landoni6, aldo mazzilli7, eugenia moretti5, lorenzo placidi8, valeria trojani2, alessandro scaggion9, claudio fiorino1

Authors Affiliations

1IRCCS San Raffaele Scientific Institute, Medical Physics, Milano, Italy; 2Departement of Advanced Technology, IRCCS Reggio Emilia, Medical Physics, Reggio Emilia, Italy; 3University Hospital of Parma AOUP, Medical Physics, Parma, Italy; 4IRCCS Veneto Institute of Technology IOV , Medical Physics, Padova, Italy; 5University Hospital of Udine, Medical Physics, Udine, Italy; 6IRCCS Istituto Nazionale dei Tumori Regina Elena , Medical Physics, Roma, Italy; 7University Hospital of Parma, Medical Physics, Parma, Italy; 8Foundation University Hospital A. Gemelli IRCCS, Medical Physics, Roma, Italy; 9IRCCS Veneto Institute of Technology IOV, Medical Physics, Padova, Italy

Purpose or Objective

Whole breast (WB) irradiation with tangential fields (TF) is a still largely used technique in post-operative radiotherapy of breast cancer. Knowledge-based (KB) plan prediction offers large opportunities in efficiently replacing manual planning, time-consuming and prone to inter-planner variability and sub-optimal planning. The expected, relatively high, consistency in contouring may translate into models usable on a large scale. First step of this process concerns the evaluation of inter institute model’s variability.

Materials and Methods

Seven Institutions delivering WB according to national guidelines with TF manually optimized using few segments per field (1-4, with/without wedges) joined a national funded project. The first aim was to model right breast TF plans by using RapidPlan (Varian Inc.), installed in each of the participating centers. Methods to generate the models, including minimum number of pts (>70), “fake”-arc geometry definition, outlier identification/elimination criteria, internal validation were previously agreed with the coordinating center. The resulting models were then imported into a dedicated Eclipse research station (V.16.7) and tested on 12 additional pts avaiable from six of the involved Institutes (2 pts per Institute). At the time of current analysis, five models were available, while two were still ongoing. DVH prediction bands of OARs (heart, ipsilateral lung, contralateral lung, contralateral breast) were analyzed. Inter-institute variability of plan performances was assessed by considering predicted DVH averaged with respect to Institutes and their (inter-institute) SD.


Predicted DVH (normalized to the 40Gy/15 fraction scheme), calculated as the mean between DVHmin and DVHmax , showed relatively good consistency, despite some not negligible inter-institute variability was seen: in Figure 1, mean DVHs of right lung (over all 12 patients) are shown for the five models with the average value incorporating inter-institute variability.  As an example, V20 mean value for ipsilateral lung was 11.3% and inter-institute SD was 1.5%. Mean doses of OARs for all institutes and their mean and SD (quantifying inter-institute variability) are shown in Table1: as expected, inter-institute variability for contralateral OARs was very small (within 0.17Gy), while inter-institute SD for ipsilateral lung was 0.36 Gy with mean institutional values ranging between 5.1 and 6.1 Gy.




These preliminary results quantify for the first time inter-institute variability of DVH prediction models in the case of WB TF. Results are encouraging showing clear potentials for efficient incorporation of inter-institute variability on KB-model and/or benchmarking. The extension to more models and more patients, to better refine the results is warranted. This study is supported by an AIRC grant (IG23150).