Safe Reinforcement Learning for Real-time Dose-based Adaptive Radiotherapy

TECoSA industrial PhD student Kenneth Lau (Elekta), will present his work so far as a “30% seminar”.  All with an interest in the topic are welcome to join, via Zoom (https://kth-se.zoom.us/j/620751730409) or in real life on KTH Campus (Teknikringen 14, room 304, floor 3).

ABSTRACT: External beam radiotherapy treats tumors using radiation beams from outside the patient’s body. However, radiation to healthy tissues can cause side effects. Current treatment planning methods use static patient images to modulate radiation beams and do not account for stochastic patient motions, such as heartbeat and breathing. This can lead to discrepancies between the plan and actual radiation delivery. MR-Linac is a recent breakthrough that allows us to visualize a patient’s anatomy in real-time. However, traditional dose-based adaptative radiotherapy methods are too slow to compute in real-time. We propose to use reinforcement learning for real-time dose-based adaptive radiotherapy.