ESTRO 2024 - Abstract Book

S4209

Physics - Intra-fraction motion management and real-time adaptive radiotherapy

ESTRO 2024

Fig. 1 – True versus predicted tumor contours for two testing patients. (Top) abdominal tumor. (Bottom) lung tumor. Orange line denotes the maximum inhalation phase.

Tab. 1 – Comparison of SegResNet prediction with ground truth contours from observer 1 or 2 respectively, and assessment of the inter-observer variability for the testing patient cohort. Mean and standard deviation are provided for each evaluation metric.

Conclusion:

Patient-specific SegResNet models have great potential for accurate real-time 2D tumor tracking in MR-guided radiotherapy. Their development only takes a few minutes and could be done in parallel to other tasks during the pre-treatment workflow.

Keywords: Deep Learning, Tumor Tracking, MRgRT

References: