ESTRO 2024 - Abstract Book

S4458

Physics - Machine learning models and clinical applications

ESTRO 2024

The AI's prediction times of under one second highlight potential efficiency gains. Couch and gantry angles by the AI were close to those chosen by human planners. This was shown both by direct comparison of angles and dosimetric analysis of resulting plans. However, the current version of the model resulted in elevated body dose outside of the target. Ongoing work is focused on an analytical post-processing algorithm to adjust the beam angle prediction reducing body dose. A previous post-processing algorithm was demonstrated to work for liver targets 1 but needs to evolve for non-coplanar fields for brain targets. In summary, this work demonstrates the feasibility for AI to automate angle selection for complex non-coplanar fields for targets in the brain treated with protons.

Keywords: Proton therapy, Beam-angle optimization

References:

1 Kaderka et al. "Toward Automatic Beam Angle Selection for Pencil-Beam Scanning Proton Liver Treatments: A Deep Learning-Based Approach" Medical Physics 2022, 49(7):4293

1251

Digital Poster

Using support vector machines (SVMs) to predict adverse outcomes in spinal SBRT

Bohdan Bodnar 1 , Kerstin Rubarth 2,3 , Goda Kalinauskaite 1 , Marcel Nachbar 1 , David Kaul 1 , Peter Vajkoczy 4 , Daniel Zips 1 , Güliz Acker 1,4,5 , Carolin Senger 1 1 Charité-Universitätsmedizin Berlin (Corporate Member of Freie Universität Berlin, Humboldt-Universität zu Berlin, and Berlin Institute of Health), Department of Radiation Oncology and Radiotherapy, Berlin, Germany. 2 Charité Universitätsmedizin Berlin (Corporate Member of Freie Universität Berlin, Humboldt-Universität zu Berlin, and Berlin Institute of Health), Institute of Biometry and Clinical Epidemiology, Berlin, Germany. 3 Charité-Universitätsmedizin Berlin (Corporate Member of Freie Universität Berlin, Humboldt-Universität zu Berlin, and Berlin Institute of Health), Institute of Medical Informatics, Berlin, Germany. 4 Charité-Universitätsmedizin Berlin (Corporate Member of Freie Universität Berlin, Humboldt-Universität zu Berlin, and Berlin Institute of Health), Department of Neurosurgery, Berlin, Germany. 5 Berlin Institute of Health at Charité - Universitätsmedizin Berlin, BIH Academy, Clinician Scientist Program, Berlin, Germany

Purpose/Objective:

Spinal metastases affect up to 30% of patients with solid tumors (1). Radiotherapy, especially stereotactic body radiation therapy (SBRT), is a well-established treatment option for those patients, either by itself or in a post operative approach. SVMs are a model of supervised learning which were developed to solve dichotomous classification problems and were already successfully implemented in medical research for outcome prediction (2). Thus, in this study, we aimed to use SVM models to predict local failures and vertebral body fractures in our cohort and to compare the models’ performance to logit models.