Online

ESTRO 2020

Session Item

Physics track: Treatment planning: applications
9322
Poster
Physics
00:00 - 00:00
Evaluation of an artificial intelligence driven planning system for online adaptive radiotherapy
Lucie Calmels, France
PO-1483

Abstract

Evaluation of an artificial intelligence driven planning system for online adaptive radiotherapy
Authors: Lina Andersson.(Herlev hospital, University of Copenhagen- Radiotherapy Research Unit, Herlev, Denmark), Lucie Calmels.(Herlev hospital, University of Copenhagen- Radiotherapy Research Unit, Herlev, Denmark), Patrik Sibolt.(Herlev hospital, University of Copenhagen- Radiotherapy Research Unit, Herlev, Denmark), Maria Sjölin.(Herlev hospital, University of Copenhagen- Radiotherapy Research Unit, Herlev, Denmark), David Sjöström .(Herlev hospital, University of Copenhagen- Radiotherapy Research Unit, Herlev, Denmark)
Show Affiliations
Purpose or Objective

Treatment planning has become more sophisticated over the past decade enabling creation of complex, dynamic radiotherapy plans with conform dose distributions, securing target coverage while sparing normal tissue. More recently artificial intelligence (AI) has been utilized to better support the planners. The aim of this study was to compare the dosimetric performance of intensity modulated radiation therapy (IMRT) and volumetric modulated arc therapy (VMAT) plans generated with both AI driven treatment planning system (TPS), developed for online adaptive radiotherapy, and standard TPS for different treatment sites.

Material and Methods

The TPS for AI-driven automatic generation of radiotherapy treatment plans uses clinical goals and dose constraints to build the underlying optimization objective functions. These sets of planning directives were, for a range of treatment sites, optimized and saved as templates in order to ensure highest rate of fulfilled goals. Systematic generation of two IMRT (9 and 12 field) and two VMAT (2 and 3 arc) treatment plans, in an emulator with a pre-clinical version of the AI-driven TPS, was based on those same templates. The reference treatment plans were correspondingly optimized to achieve the best level of target coverage with an optimal sparing of organs at risk (OARs). The automatically generated plans were transferred from the AI driven TPS to the standard TPS and compared to the reference plan. The Dose-volumetric data for the planning target volume (PTV) and for the relevant OARs, the homogeneity index (HI), the conformity index (CI), the modulation factor in terms of number of Monitor Units (MU) per Gy, and the dose to the normal tissue of both treatment approaches were all compared.

Results



 





The IMRT plans generated by the AI-driven TPS were better at fulfilling the clinical goals than the generated VMAT plans for all the treatment sites. Both, the coverage of the PTV and the sparing of OAR were very similar between the best AI-driven IMRT plans and the reference plans from the standard TPS (Table 1). The CI was similar between the two TPS for IMRT plans and better for VMAT reference plan compared to the AI driven plan. The HI was closer to 0 for the reference plan for both IMRT and VMAT plan. The modulation complexity was higher for the AI-driven VMAT plans compared to the reference plan and comparable for the IMRT treatment plans generated by the two TPS. The dose to the normal tissue was equivalent for all techniques investigated (Table 2).

Conclusion

At treatment planning level, the study demonstrated that the use of an AI-driven TPS are competitive with the current IMRT/VMAT plans created with a standard TPS and can improve the sparing of healthy tissues while maintaining a full coverage of the PTV. Further studies are required to confirm these early results.