ESTRO 2025 - Abstract Book

S3413

Physics - Machine learning models and clinical applications

ESTRO 2025

Conclusion: Our image representation offers a complete encoding of treatment plans, capturing their inherent complexity, which convolutional networks can efficiently process. Our model demonstrates robustness and adaptability to new data, proving valuable for rapidly identifying potential failures in radiotherapy plans. This could save time and resources, allowing for in-depth investigation of failure causes and supporting optimization of plan deliverability. Proper validation could make our model a beneficial pre-trained tool for other institutions, allowing them to customize it using a transfer-learning approach.

Keywords: Plan complexity, PSQA, CNN