A machine learning tool for re-planning and adaptive RT: A multicenter cohort investigation

Area of Science:

• Radiation Oncology
• Medical Imaging
• Machine Learning

Background:

• Adaptive radiotherapy (ART) aims to optimize treatment by adjusting plans based on anatomical and dosimetric changes.
• Predicting patient benefit from ART and identifying optimal re-planning times remain challenges.

Purpose of the Study:

• To develop and validate a machine learning (ML) tool for predicting patient benefit from adaptive radiotherapy (ART).
• To identify the ideal timing for re-planning interventions in head and neck cancer patients undergoing ART.

Main Methods:

• A multicenter dataset of 90 head and neck cancer patients was used, with 41 for training and 49 for testing.
• A custom ML classifier using Support Vector Machines (SVM) analyzed parotid gland (PG) volume and dose variations via deformable image registration (DIR).
• Radiation oncologists validated the ML tool's re-planning time recommendations through a double-blind evaluation.

Main Results:

• Parotid gland volume reduction averaged 23.7±8.8%.
• While 86.7% of patients did not require re-planning within the first 3 weeks, 58% showed potential benefit from the 4th week onwards.
• The 4th week was identified as the most favorable time for re-planning in 70% of cases, aligning with radiation oncologists' recommendations.

Conclusions:

• A SVM-based decision-making tool effectively addresses ART challenges.
• The study successfully identified patients who benefit from ART and determined the optimal timing for re-planning interventions.