Multimodellvalidierung von Modellen des Maschinellen Lernens von Strahlentherapien

Details

Veröffentlichungs-Nr.

EP4769423

Anmeldetag

16. Dezember 2025

Veröffentlichung

1. Juli 2026

Rechtsraum

EP

Abstract

Disclosed herein are methods (200) and systems (100) for optimizing radiation therapy plans through a dual-model system, enhancing operational efficiency, effectiveness, and safety of machine learning models (111, 112) configured to generate radiotherapy treatment plans. Embodiments utilize a utility function to guide the optimization of treatment parameters, while an independent validation model (111) evaluates each iteration of the plan. The validation model (111) ensures that the treatment remains clinically viable by halting the optimization process (200) once the validation score ceases to improve, thereby preventing over-optimization. This approach allows for optimized treatment plans to be predicted by existing machine learning in a manner that is both theoretically ideal but also practically applicable and safe for patient treatment, addressing the critical need for reliability in radiation therapy planning.

Anmelder

Firma

Siemens Healthineers International AG

Land

🇨🇭 Schweiz