ESTRO 2024 - Abstract Book

S4504

Physics - Machine learning models and clinical applications

ESTRO 2024

Conclusion:

This study shows that DL IMPT plan quality can be continuously improved after clinical introduction, through model configuration. Our study successfully updated our oropharynx DL model, by identifying and implementing common post-processing enhancements, resulting in improved plan quality for all 20 patients. This efficient update process, using a limited dataset over a short time, proves valuable for AI model sustainability in clinical practice.

Keywords: automation, head and neck cancer

1905

Digital Poster

Clinical implementation of deep learning-based dose prediction for automated planning

Janne Heikkilä 1 , Henri Korkalainen 1 , Akseli Leino 1 , Johannes Ahlnäs 2 , Minna Kauppila 2 , Arthur Sinimyrsky 2 , Juuso TJ Honkanen 1 , Jan Seppälä 1 , Tuomas Virén 1 1 Kuopio University Hospital, Department of Radiotherapy, Kuopio, Finland. 2 Kymenlaakso/Kotka Central Hospital, Department of Medical Physics, Kotka, Finland

Purpose/Objective:

Deep learning-based dose prediction is a promising technique to improve treatment plan quality and consistency as well as guide automated radiotherapy (RT) treatment planning. Here, we developed and evaluated a clinical implementation of deep learning-based dose prediction for breast cancer treatments. Furthermore, the aim was to verify that the predictions were feasible and achievable by a clinical treatment planning system. To evaluate the generalizability of the treatment planning approach, clinical data from another RT center was exploited.