ESTRO 2024 - Abstract Book

S3528

Physics - Dose prediction, optimisation and applications of photon and electron planning

ESTRO 2024

References:

1) M. Murr, K. K. Brock, M. Fusella, N. Hardcastle, M. Hussein, M. G. Jameson, I. Wahlstedt, J. Yuen, J. R. McClelland, and E. Vasquez Osorio. Applicability and usage of dose mapping/accumulation in radiotherapy. Radiotherapy and Oncology, 182:109527, 5 2023.

2) O. Weistrand and S. Svensson. The ANACONDA algorithm for deformable image registration in radiotherapy. Medical Physics, 42(1), 2015.

3) K. K. Brock, S. Mutic, T. R. McNutt, H. Li, and M. L. Kessler. Use of image registration and fusion algorithms and techniques in radiotherapy: Report of the AAPM Radiation Therapy Committee Task Group No. 132: Report. Medical Physics, 44(7), 2017.

1007

Proffered Paper

One-Click Fully Automated Treatment Planning via Deep Learning Outside a Treatment Planning System

Gerd Heilemann, Lukas Zimmermann, Wolfgang Lechner, Gregor Goldner, Joachim Widder, Dietmar Georg, Peter Kuess

Medical University of Vienna, Department of Radiation Oncology, Vienna, Austria

Purpose/Objective:

Modern radiation oncology aims to provide personalized treatments by responding to any anatomical changes in patients in real time. This requires a high level of automation and speed in the treatment planning process, which current solutions are not able to deliver. We propose a novel proof-of-concept of a fully automated one-click planning pipeline that requires no human interaction and produces a high-quality treatment plan outside a treatment planning system (TPS).

Material/Methods:

We developed an integrated pipeline that assembles a dose prediction model with a cutting-edge method to produce deliverable DICOM RT plans. Auto-segmentation, using ArtPlan (Therapanacea, France) was used to specify structures of organs-at-risk (OAR) and the target was defined by an experienced radiation oncologist. Our dose prediction technique employed a U-Net structure with embedded ResNet blocks. It further utilized One Cycle Learning, and feature-based losses to predict the dose (Zimmermann et al., 2021). The predicted dose was enhanced to optimize the dose distribution in and around the target. In this step, a filter was run over the masked dose matrix of the planning target volume (PTV) to homogenize the dose and cap the dose at 107% of the prescribed dose inside the PTV. Another filter lowered the dose in the adjacent OARs (i.e. rectum, bladder) by 2%. The enhanced dose matrices were then fed into the plan generation model (Heilemann et al., 2023). This model predicted the multi-leaf collimator (MLC) sequences for VMAT plans. Separate L1 losses were designed for leaf and jaw positions and monitor units. The respective machine parameters can then be stored as a DICOM RT plan file.