ESTRO 2020 Abstract book

S388 ESTRO 2020

Table 1 . Treatment planning system beam modeling parameters requested via IROC survey.

Conclusion Multiple cases were identified as having both poor treatment delivery and specific beam modeling parameter values outside of typical reported values. This work supports the dominating characteristics of beam modeling discrepancies in dose calculation errors and may serve as direct guidance for the community to further improve dose calculation accuracy. OC-0638 Multivariate log file analysis for MLC failure prediction A. Wojtasik 1 , M. Bolt 2 , C.H. Clark 2 , A. Nisbet 3 , T. Chen 1 1 University of Surrey, Chemical and Process Engineering, Guildford, United Kingdom ; 2 Royal Surrey County Hospital, Medical Physics, Guildford, United Kingdom ; 3 University College London, Department of Medical Physics & Biomedical Engineering, London, United Kingdom Purpose or Objective MLC failures are a relatively common cause of radiotherapy disruption. Where MLC trajectory log file data of a linac is available, it may be beneficial in helping to predict future MLC failure. A multivariable approach, comparing individual leaf behaviour to that of other surrounding leaves/MLC carriage could help distinguish uncharacteristic leaf behaviour from natural variation. This in turn has the potential to allow for proactive MLC maintenance, lessening the potential impact on patient treatment time. Material and Methods 2 months of trajectory log data from an operational Varian TrueBeam linac has been collected, encompassing all 538 treatments delivered. Based on the logs, the average error in each of the 130 parameters for each treatment was calculated. In order to track the changes in parameter behaviour over time, a moving window approach to the analysis was employed. The resultant dataset was then standardised via z-scoring. This was performed to minimise the impact of varying measurement units for different parameters on further analysis. A covariance matrix was then calculated for the scaled dataset. Leaves showing covariances atypical to the leaves around them are marked down as potential failures. Results The linac considered has had 3 reported MLC failures over the time window for which data has been analysed. All 3 of those failures were sudden, with no abnormal positioning prior to the failures being noticed in the original log files. Typical covariance of a leaf with other leaves/corresponding leaf carriage was noted to be in the 0.6-0.8 range showing very strong relationships between the variables. In the lead up to leaf failure, the covariance values for the leaf of interest drop by a noticeable amount (10-15%) 2 days prior to failure. For covariance matrices

Results 114 phantom irradiations were identified as having poor dose agreement, with 82 of these cases having complete survey responses and no setup errors. 42 (51.2%) of the examined cases had at least one variable outside the 10 th or 90 th percentiles of survey results (see Table 2). In particular, atypical DLG values for Eclipse were identified in 22 out of 33 flagged cases of poor-performing irradiations. Table 2 . Summary of reviewed phantom cases.