2023/09/11 | Research | Artificial Intelligence
Deep Learning Dose Prediction for Radiotherapy
Radiotherapy treatment requires accurate contouring of tumor volumes and organs at risk. In a collaboration between the ARTORG Center for Biomedical Engineering Research, Uni Bern, and the Department of Radiation Therapy, Inselspital, a deep learning architecture for automatic segmentation and 3D dose prediction for glioblastoma patients has shown good sensitivity and robustness.
Dosimetric comparison of the calculated dose, the initial prediction model, and the updated model for a concave (above) and a multiple lesion case (below).
(https://doi.org/10.3390/cancers15174226)
The study showed that obtaining a dose prediction model for glioblastoma VMAT (volumetric modulated arc therapy) treatment that is both robust and sensitive to realistic segmentation variations is possible with less than 100 cases. Although the model is only valid for one specific treatment strategy, the relative ease of implementation and the near-instant inference time make it a useful technique to incorporate dose awareness into the automation processes in radiotherapy prior to planning.
Link to the study
Medical Image Analysis
