ESTRO 2023 - Abstract Book

S1367

Digital Posters

ESTRO 2023

These predicted CTVs showed high correlations with manual contours. Deep Learning-based auto-segmentation models achieve acceptable accuracy and improve clinical efficiency for radiotherapy of diverse tumor types.

PO-1665 When no news is good news: commercial automated EPID in vivo dosimetry deployed as a safety check

R. Caines 1 , M. Gilmore 1

1 The Clatterbridge Cancer Centre NHS Foundation Trust, Medical Physics, Liverpool, United Kingdom

Purpose or Objective In vivo dosimetry (IVD) in radiotherapy is a strong safety recommendation within the UK and a legal requirement in many European countries. Despite this, large-scale deployment of a service-wide IVD programme is challenging in terms of required physics resource associated with calibration and maintenance of dosimetry equipment and troubleshooting out of-tolerance results. Moreover, traditional point-based measurement devices are becoming obsolete as VMAT and other rotational techniques are increasingly utilised. We describe our experience implementing and sustaining a semi-automated EPID in vivo dosimetry service as a gross safety check within a large radiotherapy centre using the Sun Nuclear SunCHECK system. Materials and Methods Following initial configuration in SunCHECK, 10 Varian linacs (7 x TrueBeam and 3 x 2100 series with aSi1000 or aSi1200 MV panels) were calibrated (July-September 2019) for acquisition of 2D integrated dosimetry images, supporting a patient load of approximately 300 new starts per month across a range of tumour sites and techniques. Subsequently as older linacs were replaced, new machines were calibrated as part of standard pre-clinical commissioning. Since implementation, patient measurements were acquired as part of routine treatment delivery, against standard exclusion criteria such as field limit (determined with an automated script within the TPS) or collision risk. Patients were typically measured at first fraction or near beginning of treatment, and each measured image (1-6 treatment fields) analysed using 2D global gamma analysis at 20% threshold. Analysis was carried out simultaneously at 10% 5 mm 'Good', 5% 5 mm 'Better', and 3% 3 mm 'Best', but deemed acceptable at >90% passing rate at 'Good' reflecting local policy and regulatory requirement in respect of significant treatment errors. (All patients also received daily IGRT). DICOM query retrieve and image processing was entirely automated within SunCHECK. Human review and approval of results was carried out daily by the local dosimetry team, with escalation of anomalous results to senior physicists following first-line troubleshooting. Results Over three years, 9440 patients were measured comprising 22,532 individual beam measurements. Mean beam gamma passing rate at 10% 5 mm was 96.0 ± 14.6%, with 20,961 fields passing local criteria (93.0%). first line dosimetry review comprised approximately 0.5 – 1 person-hours per day. Common failure modes such as calibration drift, off-axis EPID placement or beam interruptions were easy to identify and correct with follow up measurement. Second-line physics activity, including Medical Physics Expert review, typically comprised 1 person-hour per month. Four patients (0.04%) required further investigation beyond routine troubleshooting protocols. No significant treatment errors were detected. Conclusion The SunCHECK system offers an efficient and practical solution for large-scale deployment of IVD as a safety check within a VMAT-first treatment paradigm. Purpose or Objective To present our experience with an automated approach to IMRT treatment planning using expedited constrained hierarchical optimization (ECHO) for various disease sites including paraspinal/metastatic tumor, prostate and non-small cell lung cancer. Materials and Methods After contouring, a template using multiple IMRT fields was created and sent to ECHO through the treatment planning system (TPS) application program interface (API) plug-in. ECHO does not use the TPS optimization engine, however, it utilizes TPS leaf sequencing and dose calculation. Institutional clinical criteria, including maximum and mean doses plus dose volume metrics, were directly employed as template ECHO input parameters to derive appropriate objective functions and constraints. ECHO applies advanced optimization tools (such as hierarchical constrained optimization, convex approximations and Lagrangian methods) to produce Pareto optimal plans. The optimal fluence map generated by ECHO was imported automatically into TPS for leaf sequencing and final dose calculation. Upon ECHO completion, the planner received an email indicating the plan was ready for review. The plan was accepted by the planner if all clinical criteria were met, otherwise a limited number of parameters could be adjusted prior to another run with ECHO. Results For each disease site, a template based on the established clinical criteria (limit and guidelines) was created and ECHO only required limited representative training patient data to fine tune few optimization parameters (e.g., dose fall-off PO-1666 Automated clinical-criteria-driven optimal planning: clinical experience with over 6000 patients L. Hong 1 , Y. Zhou 1 , Q. Huang 1 , G. Jhanwar 1 , J. Yang 1 , H. Pham 1 , L. Cervino 1 , J. Deasy 1 , M. Zarepisheh 1 1 Memorial Sloan Kettering Cancer Center, Medical Physics, New York, USA