ESTRO 2023 - Abstract Book

S864

Digital Posters

ESTRO 2023

probabilities (NTCP) models in a comparison between e.g. proton therapy (PBS) and conventional radiotherapy, such as VMAT.

Materials and Methods PARROT is organized in four screens:

Study Management: This screen allows users to select the patient and study they want to work on. It actions an Orthanc daemon, a lightweight and standalone DICOM server. Patient data can also be directly imported from this screen and is stored locally, to ensure the confidentiality of patient files. AI Management: Users can select AI models from a list we made available. As a first proof of concept, we provide models to delineate organs at risk and to predict the dose delivered by a VMAT and PBS treatment for head and neck cancers. Users can also choose to load their own trained models for other locations by using a provided python scripting environment. Patient Modeling: Users can visualize the contours delineated by the AI model and edit them interactively. In the lower part of the screen, an history of the edition is displayed and users can navigate through previous modifications and optionally delete them. Additionally, descriptive labels to report the reason for the alterations are collected to further improve the AI models. Plan Evaluation: Two dose distributions are displayed on this screen. Dose distributions will either come from the clinical study or from the AI models. We propose several tools to compare the two dose distributions: DVH curves, other dose statistics with several DVH metrics, a comparison of clinical goals met by the different doses and a treatment indication according to a NTCP protocol. Clinical goals and NTCP models can be selected by the user.

Results Figure 1 illustrates the AI prediction workflow in PARROT and Figure 2 shows a screen of the application.

Conclusion PARROT offers a user-friendly interface to interact with AI models, display the results, edit the contours and decision support for treatment selection by displaying the comparison of predicted doses. It can be a support for the coalitional treatment planning mixing expertise and data from a network of hospitals.

PO-1080 Evaluation of the nutritional pathway for patients at risk of malnutrition undergoing radiotherapy

V. Forte 1 , F.I. Pancione 1 , V. De Simone 1 , L. Sabatino 1 , D. Di Minico 1 , S. Caponigro 1 , F. Cavallo 1 , S. Borrelli 1 , I. Russo 1 , G. Abate 1 , G. Russo 1 , N. Gennuso 1 , P. Gentile 1,2 1 UPMC Villa Maria Hillman Cancer Center, Radiotherapy, Mirabella Eclano (AV), Italy; 2 UPMC San Pietro FBF, Radiotherapy, Roma, Italy

Purpose or Objective