Session Item

To implement a Knowledge-Based (KB) automatic planning approach for left-sided whole breast treatment for the TomoDirect (TD) modality of Tomotherapy.

69 TD clinical plans were selected to train a KB-model for left sided breast cancer patients using the RapidPlan (RP) tool implemented in the Eclipse TPS (v13.6, Varian Inc). Using the association between dosimetry and anatomy/geometry features, RP generates models for accurate DVH prediction that can be used to automatize plan optimization. As RP works with inverse-planning techniques (like RapidArc, RA), virtual RA-plans were associated to the dose distributions of TD clinical plans, imported from the Tomotherapy planning station (TomoHD system v.2.1.4). The resulting KB-model was used to set an optimized tomotherapy template via Eclipse Scripting API. The template was fine-tuned to optimize pitch, field width, modulation factor, structure importance, objectives position and priority, resulting in individually KB-optimized template. 30 out of 69 patients and 10 new patients were used for internal and external validation respectively. KB fully-automatic TD plans (KB-TD) were generated for the validation tests using the same gantry entrance/number of fields of the corresponding clinical plan. The comparison between KB-TD and the original plans was made in terms of OARs/PTV dose-volume parameters. Wilcoxon-tests were performed to assess statistically significant differences (p<0.05).

9 out of 69 patients were excluded from the model, considered to be dosimetric outliers for at least one structure. In the internal validation PTV, coverage (V95%) was improved (1% p<0.05) alongside D1% (0.4Gy p<0.05) and homogeneity (dose SD) (0.09Gy p<0.05). OAR mean dose was improved (p<0.05) with the automatic approach by 0.1/0.2/0.1/0.06 Gy for ipsil. lung/heart/contral. breast/ contral. lung respectively. 4 plans out of 30 resulted to be unacceptable in terms of PTV coverage: manually inserting one/two beams (up to a maximum of four) was sufficient to make the plans acceptable. Concerning the external validation, the general trend was replicated even though some differences were found to be not significant (p>0.05), due to the small available sample (Tab.1-Fig.1). All 10 plans were considered acceptable. A field width of 2.5cm was used for all KB-TD plans, resulting in a delivery time of 8±1min, comparing well with our clinical experience.

We proposed an approach for large-scale automatic planning using the commercially available RP tool from Varian in a different environment. The KB-TD approach was able to generate automatic plans comparable to the clinical ones or slightly better in terms of PTV coverage, PTV dose homogeneity and Dm to body and OARs. Using four beams, no manual adjustment is needed after automatic optimization. The suggested approach is under clinical implementation, fully replacing manual plan optimization for left whole-breast tomotherapy with TD. The study is supported by an AIRC grant (IG23150)