S2250
Physics - Intra-fraction motion management and real-time adaptive radiotherapy
ESTRO 2026
1 Image X Institute, University of Sydney, Sydney, Australia. 2 Central Coast Cancer Centre, Central Coast Local Health District, Gosford, Australia Purpose/Objective: 71% of the ESTRO community want increased real- time IGRT,1due to compelling clinical drivers including the MIRAGE phase 3 trial2 demonstrating a 50% toxicity reduction. However, the highest barriers to adoption are resources and equipment.1 To solve this problem, an AI markerless real-time IGRTsystem3,4has been developed that can be delivered on standard linacs. The goal was to investigate the AI markerless real-time IGRTsystem for liver cancer SABR. Two hypotheses from the NKI Master Trial5 were adopted.H1, reliability:The proportion of completed- to-intended fractions is 1.0.H2, performance:Absence of geographical miss in ≥ 85% of fractions. Material/Methods: Five liver cancer patients treated with SABR enrolled in theLearning Environment for Artificial Intelligence in Radiotherapy New Technology (LEARN) study(NCT05184790) were included. The patients had biocompatible polymer markers implanted near the tumour. A cGAN- based AI markerless real-time IGRTsystem3,4 was deployed parallel to the patient treatment for two fractions for each patient (Figure 1). The AI system was not used to make any treatment decisions, a requirement of the LEARN study. The tracking error was quantified by the difference between the AI-estimated GTV centroid position and the marker-inferred GTV centroid position.H1, reliability, was tested using binomial statistics. H2, performance, was quantified bymeasuring the proportion of fractions in which tracking error was less than the CTV to PTV margin, 5 mm,for 85% of the fraction duration. H2 was tested using binomial statistics. The overall accuracy of the AI markerless real-time IGRT system was quantified by the mean and standard deviation of the tracking error. The system latency was the sum of the time forimage reading and processing. Results: The AI markerless real-time IGRTsystem was used in 10 completed of 10 intended fractions, a 1.0 proportion, 95% confidence limits [0.69,1.00], consistent with the reliability hypothesis, H1. The absence of geographical misswas observed in 5 of 10 fractions, a 0.5 proportion, 95% confidence limits [0.19,0.81], inconsistent with the performance hypothesis, H2.The accuracy for all patients (Figure 2) was0.3±2.0mm in the superior-inferior direction and 0.0±1.9mm in the AP-lateral direction. The system latency was 195±24ms.
Conclusion: MANIV-DIBH enables highly reproducible breath-holds during RT with mean deviations below 1 mm. This analysis supports the use of reduced PTV margins (5 mm), improving treatment precision in RT. References: 1. Van Ooteghem et al., Mechanically-assisted and non-invasive ventilation for radiation therapy: A safe technique to regularize and modulate internal tumour motion. Radiother Oncol, 2019. 141: p 283-2912. Vander Veken et al., Voluntary versus mechanically- induced deep inspiration breath-hold for left breast cancer: A randomized controlled trial. Radiother Oncol, 2023. 183: p. 109598.3. Pierrard et al., Mechanically assisted non-invasive ventilation for liver SABR: Improve CBCT, treat more accurately. Clin Transl Radiat Oncol, 2025. 53: p 1009834. Van Herk et al., The probability of correct target dosage: dose-population histograms for deriving treatment margins in radiotherapy. Int J Radiat Oncol Biol Phys, 2000. 47(4): p 1121-35 Keywords: MANIV-DIBH, Intrafraction Deviations, Mobile Tumor
Digital Poster 3495
First-in-Human AI markerless real-time IGRT for liver SABR results from the Learning Environment for AI in Radiotherapy New Technology (LEARN) study Gregory Willson 1 , Freeman Jin 1 , Benjamin Zwan 2 , Iliana Peters 2 , Chandrima Sengupta 1 , Paul Keall 1
Made with FlippingBook - Share PDF online